Loading video...

Video Failed to Load

Go Home

New LLMs that control UIs! ByteDance Research releases UI-TARS, fine-tuned GUI agent that integrates reasoning, and action capabilities into a single vision-language model. Think of computer use but open. 👀 TL;DR; 3️⃣ Available in 3 sizes: 2B, 7B, and 72B parameters 🧠 Trained Qwen2-VL models with SFT & DPO...

_philschmid's profile picture

Philipp Schmid

101,465 subscribers

48,170 views • 1 year ago •via X (Twitter)

Science & TechnologyNews & Politics
LIVE

Anya Rossi• Live Now

Private livecam show

2 Comments

Philipp Schmid's profile picture
Philipp Schmid1 year ago

Code: Paper: Models: Deskop App:

UNPLUGGED PERFORMANCE's profile picture
UNPLUGGED PERFORMANCE2 years ago

Upgrade your Tesla with UP-03 Forged Wheels from Unplugged Performance! Unmatched strength, lightweight design, and track-proven durability—perfect for Model S, 3, X, and Y. Ready to ship with a lifetime warranty. #Tesla #UP03 #UnpluggedPerformance

Related Videos

Molmo by Ai2 - Open source SoTA Multimodal (Vision) Language model, beating Claude 3.5 Sonnet, GPT4V and comparable to GPT4o 🔥 They release four model checkpoints: 1. MolmoE-1B, a mixture of experts model with 1B (active) 7B (total) 2. Molmo-7B-O, most open 7B model 3. Molmo-7B-D, demo model 4. Molmo-72B, best model System Architecture > Input: Multi-scale, multi-crop images generated from the original image. > Vision Encoder: OpenAI's ViT-L/14 336px CLIP model, a powerful ViT, encodes images into vision tokens. > Connector: MLP projects tokens to LLM input space, followed by pooling for dimensionality reduction. > LLM: Decoder-only Transformer, various options (OLMo, OLMoE, Qwen2, Mistral, Gemma2, Phi) with diverse scales and openness. Model Variants > Vision Encoder: Consistent ViT-L/14 CLIP model across variants. > LLM: OLMo-7B-1024, OLMoE-1B-7B-0924, Qwen2 (7B, 72B), Mistral 7B, Gemma2 9B, Phi 3 Medium, offering different capacities and openness levels. Training Strategy > Stage 1: Multimodal pre-training for caption generation with new captioning data. > Stage 2: Supervised fine-tuning on a dataset mixture, updating all parameters. > No RLHF involved, Learning rates adjusted based on component types and pre-training status. > All the weights are available on Hugging Face Hub 🤗 > Compatible with Transformers (Remote Code) Kudos Ai2 for such a brilliant and open work! 🐐 Video credits: Allen AI YT Channel
2:15

Molmo by Ai2 - Open source SoTA Multimodal (Vision) Language model, beating Claude 3.5 Sonnet, GPT4V and comparable to GPT4o 🔥 They release four model checkpoints: 1. MolmoE-1B, a mixture of experts model with 1B (active) 7B (total) 2. Molmo-7B-O, most open 7B model 3. Molmo-7B-D, demo model 4. Molmo-72B, best model System Architecture > Input: Multi-scale, multi-crop images generated from the original image. > Vision Encoder: OpenAI's ViT-L/14 336px CLIP model, a powerful ViT, encodes images into vision tokens. > Connector: MLP projects tokens to LLM input space, followed by pooling for dimensionality reduction. > LLM: Decoder-only Transformer, various options (OLMo, OLMoE, Qwen2, Mistral, Gemma2, Phi) with diverse scales and openness. Model Variants > Vision Encoder: Consistent ViT-L/14 CLIP model across variants. > LLM: OLMo-7B-1024, OLMoE-1B-7B-0924, Qwen2 (7B, 72B), Mistral 7B, Gemma2 9B, Phi 3 Medium, offering different capacities and openness levels. Training Strategy > Stage 1: Multimodal pre-training for caption generation with new captioning data. > Stage 2: Supervised fine-tuning on a dataset mixture, updating all parameters. > No RLHF involved, Learning rates adjusted based on component types and pre-training status. > All the weights are available on Hugging Face Hub 🤗 > Compatible with Transformers (Remote Code) Kudos Ai2 for such a brilliant and open work! 🐐 Video credits: Allen AI YT Channel

Vaibhav (VB) Srivastav

80,474 views • 1 year ago

Nvidia has just announced Alpamayo 2 Super, an open 34 billion parameter reasoning vision-language-action model designed to accelerate the development of autonomous vehicles. This new model combines the NVIDIA Cosmos 3 Super reasoning model with a 2 billion parameter diffusion-based action expert model, and is post trained with reinforcement learning. The model can return multiple outputs: future trajectory plans, reasoning traces, grounded answers to questions about the scenes, and auto label generation. The model weights are now available for anyone to download on Hugging Face, and the inference code has been posted to GitHub. Distilled models can be deployed commercially without any further permission from Nvidia, and model outputs carry no license conditions. Automakers can distill down a compact version of this model that can run on the Nvidia computer in the car. Major kudos to Nvidia and Jensen Huang for advancing the state of the industry by releasing this as an open model with permissive licensing. Jensen isn't just paying lip service to the idea of open models, Nvidia is actually contributing to the ecosystem — and it's great for their business, because it helps sell more Thor computers that go in the car. Anyone can go download the model and play with it. If you do, let me know what you think. Personally I think it's so cool that we have open weights models that are this advanced, for anyone to download.
2:30

Nvidia has just announced Alpamayo 2 Super, an open 34 billion parameter reasoning vision-language-action model designed to accelerate the development of autonomous vehicles. This new model combines the NVIDIA Cosmos 3 Super reasoning model with a 2 billion parameter diffusion-based action expert model, and is post trained with reinforcement learning. The model can return multiple outputs: future trajectory plans, reasoning traces, grounded answers to questions about the scenes, and auto label generation. The model weights are now available for anyone to download on Hugging Face, and the inference code has been posted to GitHub. Distilled models can be deployed commercially without any further permission from Nvidia, and model outputs carry no license conditions. Automakers can distill down a compact version of this model that can run on the Nvidia computer in the car. Major kudos to Nvidia and Jensen Huang for advancing the state of the industry by releasing this as an open model with permissive licensing. Jensen isn't just paying lip service to the idea of open models, Nvidia is actually contributing to the ecosystem — and it's great for their business, because it helps sell more Thor computers that go in the car. Anyone can go download the model and play with it. If you do, let me know what you think. Personally I think it's so cool that we have open weights models that are this advanced, for anyone to download.

Whole Mars Catalog

45,595 views • 26 days ago

InternLM-XComposer-2.5 A Versatile Large Vision Language Model Supporting Long-Contextual Input and Output We present InternLM-XComposer-2.5 (IXC-2.5), a versatile large-vision language model that supports long-contextual input and output. IXC-2.5 excels in various text-image comprehension and composition applications, achieving GPT-4V level capabilities with merely 7B LLM backend. Trained with 24K interleaved image-text contexts, it can seamlessly extend to 96K long contexts via RoPE extrapolation. This long-context capability allows IXC-2.5 to excel in tasks requiring extensive input and output contexts. Compared to its previous 2.0 version, InternLM-XComposer-2.5 features three major upgrades in vision-language comprehension: (1) Ultra-High Resolution Understanding, (2) Fine-Grained Video Understanding, and (3) Multi-Turn Multi-Image Dialogue. In addition to comprehension, IXC-2.5 extends to two compelling applications using extra LoRA parameters for text-image composition: (1) Crafting Webpages and (2) Composing High-Quality Text-Image Articles. IXC-2.5 has been evaluated on 28 benchmarks, outperforming existing open-source state-of-the-art models on 16 benchmarks. It also surpasses or competes closely with GPT-4V and Gemini Pro on 16 key tasks.
2:45

InternLM-XComposer-2.5 A Versatile Large Vision Language Model Supporting Long-Contextual Input and Output We present InternLM-XComposer-2.5 (IXC-2.5), a versatile large-vision language model that supports long-contextual input and output. IXC-2.5 excels in various text-image comprehension and composition applications, achieving GPT-4V level capabilities with merely 7B LLM backend. Trained with 24K interleaved image-text contexts, it can seamlessly extend to 96K long contexts via RoPE extrapolation. This long-context capability allows IXC-2.5 to excel in tasks requiring extensive input and output contexts. Compared to its previous 2.0 version, InternLM-XComposer-2.5 features three major upgrades in vision-language comprehension: (1) Ultra-High Resolution Understanding, (2) Fine-Grained Video Understanding, and (3) Multi-Turn Multi-Image Dialogue. In addition to comprehension, IXC-2.5 extends to two compelling applications using extra LoRA parameters for text-image composition: (1) Crafting Webpages and (2) Composing High-Quality Text-Image Articles. IXC-2.5 has been evaluated on 28 benchmarks, outperforming existing open-source state-of-the-art models on 16 benchmarks. It also surpasses or competes closely with GPT-4V and Gemini Pro on 16 key tasks.

AK

66,472 views • 2 years ago

Introducing PhoneLLM, an open model for voice agents. GPT 5.6 Terra performance on typical voice agent tasks at 1/3 the latency and 1/18 the cost. For voice agents, we need models that are both very low latency and very good at tool calling and instruction following. There's a trade-off here, and we often have to compromise on either latency or capability when building voice agents. With PhoneLLM (and the training and data stack that made this model possible) we're fixing this problem. For the last couple of years, most of the effort in frontier model development has gone towards leveraging test-time compute. Which is awesome! Models of all shapes and sizes are available that perform really, really well ... if you have "thinking" turned on for your model. But if you need your agent to respond at voice conversation speed, you can't use thinking models. PhoneLLM is a full-weights fine-tune of NVIDIA Nemotron Nano 30B. We trained on a wide range of real-world telephone and customer support use cases. The training focused on taking the excellent Nano 30B base capabilities and teaching the model to do typical voice agent tasks with thinking disabled. The results are really good: accurate tool calling and concise, on-topic responses in long conversations. And fast: TTFAT measured server-side is <100ms if you run PhoneLLM on a lightly loaded B200. :-) But seriously, when we characterize model latency, we do it with full, end-to-end, batched request simulations using real Pipecat voice agent pipelines. You can serve more than 80 concurrent agents on a single B200 with P95 end-to-end TTFAT <600ms. Including network overhead. That's an LLM cost-per-minute around $0.0025. (1/4 of a cent.) At a latency lower than any third-party API offers today. More details about this model, including weights on Hugging Face, how to spin it up with one click on Modal, and a starter project repo you can clone, are in the thread ...
3:28

Introducing PhoneLLM, an open model for voice agents. GPT 5.6 Terra performance on typical voice agent tasks at 1/3 the latency and 1/18 the cost. For voice agents, we need models that are both very low latency and very good at tool calling and instruction following. There's a trade-off here, and we often have to compromise on either latency or capability when building voice agents. With PhoneLLM (and the training and data stack that made this model possible) we're fixing this problem. For the last couple of years, most of the effort in frontier model development has gone towards leveraging test-time compute. Which is awesome! Models of all shapes and sizes are available that perform really, really well ... if you have "thinking" turned on for your model. But if you need your agent to respond at voice conversation speed, you can't use thinking models. PhoneLLM is a full-weights fine-tune of NVIDIA Nemotron Nano 30B. We trained on a wide range of real-world telephone and customer support use cases. The training focused on taking the excellent Nano 30B base capabilities and teaching the model to do typical voice agent tasks with thinking disabled. The results are really good: accurate tool calling and concise, on-topic responses in long conversations. And fast: TTFAT measured server-side is <100ms if you run PhoneLLM on a lightly loaded B200. :-) But seriously, when we characterize model latency, we do it with full, end-to-end, batched request simulations using real Pipecat voice agent pipelines. You can serve more than 80 concurrent agents on a single B200 with P95 end-to-end TTFAT <600ms. Including network overhead. That's an LLM cost-per-minute around $0.0025. (1/4 of a cent.) At a latency lower than any third-party API offers today. More details about this model, including weights on Hugging Face, how to spin it up with one click on Modal, and a starter project repo you can clone, are in the thread ...

kwindla

314,586 views • 3 days ago

There is a beautiful story that just happened in AI so let me share it for a lighter tone weekend post among all the doom stories in our AI field this week. It’s a story of people on three continents building and sharing in the open a new small efficient and state-of-the-art AI model. It started a couple of months ago when a new team in the AI scene released their first model from their headquarters in Paris (France): Mistral 7B. Impressive model, small and very strong performances in the benchmarks, better than all previous models of this size. And open source! So you could build on top of it. Lewis in Bern (Switzerland) and Ed (in Lyon, in the South of France) both from the H4 team, a team of researchers in model fine-tuning and alignment were talking about it over a coffee, in one of these gatherings that often happen at Hugging Face to break the distance between people (literal distance as HF is a remote company). What about fine-tuning it using this new DPO method that a research team from Stanford in California just posted on Arxiv, says one? Hey, that’s a great idea, replies the other. We've just build a great code base (with Nathan, Nazneen, Costa, Younes and all the H4 team and TRL community) let's use it! The next day they start diving in the datasets openly shared on the HF hub and stumble upon two interesting large and good quality fine-tuning datasets recently open-sourced by OpenBMB, a Chinese team from Tsinghua: UltraFeedback and UltraChat. A few rounds of training experiments confirm the intuition, the resulting model is super strong, by far the strongest they have ever seen in their benchmarks from Berkeley and Stanford (LMSYS and Alpaca). Join Clementine, the big boss of the open evaluation leaderboard. Her deep dive into the model capabilities confirms the results: impressive performance. But the H4 team also hosts a famous faculty member, Pr. Sasha Rush, Associate Professor at Cornell University in his daytime, hacker at HF in his nighttime. Joining the conversation, he proposes to quickly draft a research paper to organize and share all the details with the community. A few days later, the model, called Zephyr (a wind like Mistral), paper, and all details are shared with the world. Quickly other companies, everywhere in the world starts to use it. LlamaIndex, a famous data framework and community, shares how the model blew their expectations on real-life use-case benchmarks, while researchers and practitioners discuss the paper and work on the Hugging Face hub. All this happened in just a few weeks catalyzed by open access to knowledge, models, research, and datasets released all over the world (Europe, California, China) and by the idea that people can build upon one another work in AI to bring real-world value with efficient and open models. Stories like this are numerous everywhere around us and make me really proud of the AI community and see how we can build amazingly useful things together. [the video is just me reading this Friday post hahah]
4:30

There is a beautiful story that just happened in AI so let me share it for a lighter tone weekend post among all the doom stories in our AI field this week. It’s a story of people on three continents building and sharing in the open a new small efficient and state-of-the-art AI model. It started a couple of months ago when a new team in the AI scene released their first model from their headquarters in Paris (France): Mistral 7B. Impressive model, small and very strong performances in the benchmarks, better than all previous models of this size. And open source! So you could build on top of it. Lewis in Bern (Switzerland) and Ed (in Lyon, in the South of France) both from the H4 team, a team of researchers in model fine-tuning and alignment were talking about it over a coffee, in one of these gatherings that often happen at Hugging Face to break the distance between people (literal distance as HF is a remote company). What about fine-tuning it using this new DPO method that a research team from Stanford in California just posted on Arxiv, says one? Hey, that’s a great idea, replies the other. We've just build a great code base (with Nathan, Nazneen, Costa, Younes and all the H4 team and TRL community) let's use it! The next day they start diving in the datasets openly shared on the HF hub and stumble upon two interesting large and good quality fine-tuning datasets recently open-sourced by OpenBMB, a Chinese team from Tsinghua: UltraFeedback and UltraChat. A few rounds of training experiments confirm the intuition, the resulting model is super strong, by far the strongest they have ever seen in their benchmarks from Berkeley and Stanford (LMSYS and Alpaca). Join Clementine, the big boss of the open evaluation leaderboard. Her deep dive into the model capabilities confirms the results: impressive performance. But the H4 team also hosts a famous faculty member, Pr. Sasha Rush, Associate Professor at Cornell University in his daytime, hacker at HF in his nighttime. Joining the conversation, he proposes to quickly draft a research paper to organize and share all the details with the community. A few days later, the model, called Zephyr (a wind like Mistral), paper, and all details are shared with the world. Quickly other companies, everywhere in the world starts to use it. LlamaIndex, a famous data framework and community, shares how the model blew their expectations on real-life use-case benchmarks, while researchers and practitioners discuss the paper and work on the Hugging Face hub. All this happened in just a few weeks catalyzed by open access to knowledge, models, research, and datasets released all over the world (Europe, California, China) and by the idea that people can build upon one another work in AI to bring real-world value with efficient and open models. Stories like this are numerous everywhere around us and make me really proud of the AI community and see how we can build amazingly useful things together. [the video is just me reading this Friday post hahah]

Thomas Wolf

169,200 views • 2 years ago

We've officially released and open-sourced HunyuanImage 2.1, our latest text-to-image model. The new model delivers on our commitment to balancing performance and quality. With native 2K image generation, HunyuanImage 2.1 is an advanced open-source text-to-image model.🎨 ✨ New in 2.1: 🔹Advanced Semantics: Supports ultra-long and complex prompts of up to 1000 tokens, and precisely controls the generation of multiple subjects in a single image. 🔹Precise Chinese and English Text Rendering with seamless image–text integration: The model naturally integrates text into images, making it suitable for a wide range of applications such as product covers, illustrations, and poster design to meet the needs of various fields. 🔹Rich Styles and High Aesthetic: Capable of generating images in various styles—including photorealistic portraits, comics, and vinyl figures—it delivers outstanding visual appeal and artistic quality. 🔹High-Quality Generation: Efficiently produces ultra-high-definition (2K) images in the same time other models take to generate a 1K image. HunyuanImage 2.1 uses two text encoders: a multimodal large language model (MLLM) to improve the model's image and text alignment capabilities, and a multi-language character-aware encoder to improve text rendering capabilities. The model is a single- and double-stream diffusion transformer with 17B parameters. We've also open-sourced the weights of the the accelerated version with meanflow which reduces inference steps from 100 to just 8, and PromptEnhancer, the first industrial-grade rewriting model that enhances your prompts for more nuanced and expressive image generation. Now, creators turn complex ideas—like posters with slogans or multi-panel comics—into visuals faster than ever. We’re just getting started. Stay tuned for our native multimodal image generation model coming soon. 🌐Website: 🔗Github: 🤗Hugging Face: ✨Hugging Face Demo:
1:39

We've officially released and open-sourced HunyuanImage 2.1, our latest text-to-image model. The new model delivers on our commitment to balancing performance and quality. With native 2K image generation, HunyuanImage 2.1 is an advanced open-source text-to-image model.🎨 ✨ New in 2.1: 🔹Advanced Semantics: Supports ultra-long and complex prompts of up to 1000 tokens, and precisely controls the generation of multiple subjects in a single image. 🔹Precise Chinese and English Text Rendering with seamless image–text integration: The model naturally integrates text into images, making it suitable for a wide range of applications such as product covers, illustrations, and poster design to meet the needs of various fields. 🔹Rich Styles and High Aesthetic: Capable of generating images in various styles—including photorealistic portraits, comics, and vinyl figures—it delivers outstanding visual appeal and artistic quality. 🔹High-Quality Generation: Efficiently produces ultra-high-definition (2K) images in the same time other models take to generate a 1K image. HunyuanImage 2.1 uses two text encoders: a multimodal large language model (MLLM) to improve the model's image and text alignment capabilities, and a multi-language character-aware encoder to improve text rendering capabilities. The model is a single- and double-stream diffusion transformer with 17B parameters. We've also open-sourced the weights of the the accelerated version with meanflow which reduces inference steps from 100 to just 8, and PromptEnhancer, the first industrial-grade rewriting model that enhances your prompts for more nuanced and expressive image generation. Now, creators turn complex ideas—like posters with slogans or multi-panel comics—into visuals faster than ever. We’re just getting started. Stay tuned for our native multimodal image generation model coming soon. 🌐Website: 🔗Github: 🤗Hugging Face: ✨Hugging Face Demo:

Tencent Hy

89,257 views • 11 months ago

Alibaba just released a coding model that hits 82 percent on SWE-Bench Verified. That is the highest score ever published for an open-source model. The weights are free. The license is Apache 2.0. You can run it today. The model is Qwen 4 Coder 32B. Here is what 82 percent on SWE-Bench Verified actually means. SWE-Bench Verified tests whether an AI can autonomously resolve real bugs pulled from real production GitHub repositories. Not synthetic exercises. Real open-source projects that real teams depend on. A model gets a bug report, reads the code, writes a fix, and either passes the test suite or it does not. At 82 percent, Qwen 4 Coder 32B resolves 82 out of every 100 real production bugs it is given. Without a human guiding it. On code it has never seen before. For comparison: Qwen 4 Coder 32B: 82 percent SWE-Bench Verified. Open source. Apache 2.0. Claude Fable 5: 80.3 percent SWE-Bench Pro. $10 input / $50 output per million tokens. Currently suspended. GPT-5.6 Sol: Competitive on Terminal-Bench. $5 input / $30 output per million tokens. An open-weight model that you can download and run for free just beat both of them on the benchmark designed to measure real software engineering capability. Here is the architecture. Qwen 4 Coder 32B is a 32 billion parameter dense model. Not a Mixture-of-Experts. Every parameter is active on every request. This matters for inference: a dense 32B model runs on 22 gigabytes of VRAM, which fits on a single high-end consumer GPU or a MacBook Pro with 64GB of unified memory. The smaller variant, Qwen 4 Coder 4B, runs at approximately 135 tokens per second on an M5 Max and fits inside 8 gigabytes of RAM. For a model with usable coding capability, that is a new bar for what fits in a single laptop. The training methodology continued Alibaba's approach of reinforcement learning on verifiable coding tasks. The model gets rewarded when its code passes tests. It gets penalized when it fails. Over millions of training steps, the model learns to write code that actually runs rather than code that looks plausible. License: Apache 2.0. Full commercial use. No attribution requirement. No revenue threshold. No monthly active user ceiling. Weights: Hugging Face, available today. Runs on: vLLM, Ollama, SGLang, and any standard GGUF-compatible inference engine. Qwen 4 32B also runs at approximately 135 tokens per second on an M5 Max chip, setting a new bar for what a sub-8GB model can do on Apple Silicon. The open-source coding model just beat the best closed-source model in the world on the benchmark designed to test whether AI can actually do software engineering. The weights are free. The subscription is optional. Source: Autom8Labs AI Insight July 2026, State of Open Source LLMs June 2026, Kunal Ganglani blog June 2026.
0:25

Alibaba just released a coding model that hits 82 percent on SWE-Bench Verified. That is the highest score ever published for an open-source model. The weights are free. The license is Apache 2.0. You can run it today. The model is Qwen 4 Coder 32B. Here is what 82 percent on SWE-Bench Verified actually means. SWE-Bench Verified tests whether an AI can autonomously resolve real bugs pulled from real production GitHub repositories. Not synthetic exercises. Real open-source projects that real teams depend on. A model gets a bug report, reads the code, writes a fix, and either passes the test suite or it does not. At 82 percent, Qwen 4 Coder 32B resolves 82 out of every 100 real production bugs it is given. Without a human guiding it. On code it has never seen before. For comparison: Qwen 4 Coder 32B: 82 percent SWE-Bench Verified. Open source. Apache 2.0. Claude Fable 5: 80.3 percent SWE-Bench Pro. $10 input / $50 output per million tokens. Currently suspended. GPT-5.6 Sol: Competitive on Terminal-Bench. $5 input / $30 output per million tokens. An open-weight model that you can download and run for free just beat both of them on the benchmark designed to measure real software engineering capability. Here is the architecture. Qwen 4 Coder 32B is a 32 billion parameter dense model. Not a Mixture-of-Experts. Every parameter is active on every request. This matters for inference: a dense 32B model runs on 22 gigabytes of VRAM, which fits on a single high-end consumer GPU or a MacBook Pro with 64GB of unified memory. The smaller variant, Qwen 4 Coder 4B, runs at approximately 135 tokens per second on an M5 Max and fits inside 8 gigabytes of RAM. For a model with usable coding capability, that is a new bar for what fits in a single laptop. The training methodology continued Alibaba's approach of reinforcement learning on verifiable coding tasks. The model gets rewarded when its code passes tests. It gets penalized when it fails. Over millions of training steps, the model learns to write code that actually runs rather than code that looks plausible. License: Apache 2.0. Full commercial use. No attribution requirement. No revenue threshold. No monthly active user ceiling. Weights: Hugging Face, available today. Runs on: vLLM, Ollama, SGLang, and any standard GGUF-compatible inference engine. Qwen 4 32B also runs at approximately 135 tokens per second on an M5 Max chip, setting a new bar for what a sub-8GB model can do on Apple Silicon. The open-source coding model just beat the best closed-source model in the world on the benchmark designed to test whether AI can actually do software engineering. The weights are free. The subscription is optional. Source: Autom8Labs AI Insight July 2026, State of Open Source LLMs June 2026, Kunal Ganglani blog June 2026.

Harman

41,278 views • 1 month ago

PDF parsing is still painful because LLMs reorder text in complex layouts, break tables across pages, and fail on graphs or images. 💡Testing the new open-source OCRFlux model, and here the results are really good for a change. So OCRFlux is a multimodal, LLM based toolkit for converting PDFs and images into clean, readable, plain Markdown text. Because the underlying VLM is only 3B param, it runs even on a 3090 GPU. The model is available on Hugging Face . The engine that powers the OCRFlux, teaches the model to rebuild every page and then stitch fragments across pages into one clean Markdown file. It bundles one vision language model with 3B parameters that was fine-tuned from Qwen 2.5-VL-3B-Instruct for both page parsing and cross-page merging. OCRFlux reads raw page images and, guided by task prompts, outputs Markdown for each page and merges split elements across pages. The evaluation shows Edit Distance Similarity (EDS) 0.967 and cross‑page table Tree Edit Distance 0.950, so the parser is both accurate and layout aware. How it works while parsing each page - Convert into text with a natural reading order, even in the presence of multi-column layouts, figures, and insets - Support for complicated tables and equations - Automatically removes headers and footers Cross-page table/paragraph merging - Cross-page table merging - Cross-page paragraph merging A compact vision‑language models can beat bigger models once cross‑page context is added. 🧵 1/n Read on 👇
0:44

PDF parsing is still painful because LLMs reorder text in complex layouts, break tables across pages, and fail on graphs or images. 💡Testing the new open-source OCRFlux model, and here the results are really good for a change. So OCRFlux is a multimodal, LLM based toolkit for converting PDFs and images into clean, readable, plain Markdown text. Because the underlying VLM is only 3B param, it runs even on a 3090 GPU. The model is available on Hugging Face . The engine that powers the OCRFlux, teaches the model to rebuild every page and then stitch fragments across pages into one clean Markdown file. It bundles one vision language model with 3B parameters that was fine-tuned from Qwen 2.5-VL-3B-Instruct for both page parsing and cross-page merging. OCRFlux reads raw page images and, guided by task prompts, outputs Markdown for each page and merges split elements across pages. The evaluation shows Edit Distance Similarity (EDS) 0.967 and cross‑page table Tree Edit Distance 0.950, so the parser is both accurate and layout aware. How it works while parsing each page - Convert into text with a natural reading order, even in the presence of multi-column layouts, figures, and insets - Support for complicated tables and equations - Automatically removes headers and footers Cross-page table/paragraph merging - Cross-page table merging - Cross-page paragraph merging A compact vision‑language models can beat bigger models once cross‑page context is added. 🧵 1/n Read on 👇

Rohan Paul

149,292 views • 1 year ago

Voice AI turn taking is a solved problem. The single most common complaint about voice AI, today, is that agents interrupt too often. But the voice agents I build for myself now respond quickly and interrupt me less often than the people I talk to every day. (I actually measured this.) Mark Backman made a Pipecat AI PR two weeks ago that was the last piece of the puzzle for turn taking so good that I no longer ever think about it. The approach combines three layers of processing: 1. Voice activity detection, with a short (200ms) trigger. 2. A native audio turn detection model that's small, fast, and runs on CPU. This model captures audio nuances like inflection and filler sounds that don't get transcribed. 3. A prompt mixin for the conversation LLM that decides turn completion based on conversation context. None of these are new. We've been using VAD for a long time. We trained the first version of the Pipecat Smart Turn native audio model in December 2024. And we've been experimenting with prompt-based large model turn detection (sometimes called "selective refusal") for more than a year. Now, the Smart Turn model and the SOTA LLMs we're using in voice agents have both gotten so good that using them together feels like we've finally "solved" turn detection. Mark also figured out how to elegantly apply a "single-token tagging" technique to this problem. We sometimes use single-token tagging in place of tool calling, when we need a near-zero latency programmatic trigger. Mark's Pipecat mixin defines three single-token characters and prompts the LLM to output exactly one of them at the beginning of every response. - ✓ means the agent should respond normally (immediately) - ○ is a "short incomplete" - the agent should wait 5 seconds - ◐ is a "long incomplete" - the agent should wait 10 seconds The wait times, and the details of the prompt, are configurable, of course. Watch the video to see me talk to an agent that handles all my various pauses and inflections, plus phrases like "let me think," pretty much the way a person would handle them, in terms of response latency. Also, in the second half of the video, I ask the agent to adjust its response pattern because I'm going to tell it a phone number. This kind of "in-context" adjustment of response wait times is really useful. The LLM in the video is GTP-4.1. We've tested the prompt and single-token adherance with GPT-4.1, Gemini 2.5 Flash, Anthropic Claude Sonnet 4.5, and AWS Nova 2 Pro. Note that older models in all these families (and, in general, smaller open weights models) aren't able to reliably output these single-token tags. But the new models we're using these days are pretty amazing.
4:39

Voice AI turn taking is a solved problem. The single most common complaint about voice AI, today, is that agents interrupt too often. But the voice agents I build for myself now respond quickly and interrupt me less often than the people I talk to every day. (I actually measured this.) Mark Backman made a Pipecat AI PR two weeks ago that was the last piece of the puzzle for turn taking so good that I no longer ever think about it. The approach combines three layers of processing: 1. Voice activity detection, with a short (200ms) trigger. 2. A native audio turn detection model that's small, fast, and runs on CPU. This model captures audio nuances like inflection and filler sounds that don't get transcribed. 3. A prompt mixin for the conversation LLM that decides turn completion based on conversation context. None of these are new. We've been using VAD for a long time. We trained the first version of the Pipecat Smart Turn native audio model in December 2024. And we've been experimenting with prompt-based large model turn detection (sometimes called "selective refusal") for more than a year. Now, the Smart Turn model and the SOTA LLMs we're using in voice agents have both gotten so good that using them together feels like we've finally "solved" turn detection. Mark also figured out how to elegantly apply a "single-token tagging" technique to this problem. We sometimes use single-token tagging in place of tool calling, when we need a near-zero latency programmatic trigger. Mark's Pipecat mixin defines three single-token characters and prompts the LLM to output exactly one of them at the beginning of every response. - ✓ means the agent should respond normally (immediately) - ○ is a "short incomplete" - the agent should wait 5 seconds - ◐ is a "long incomplete" - the agent should wait 10 seconds The wait times, and the details of the prompt, are configurable, of course. Watch the video to see me talk to an agent that handles all my various pauses and inflections, plus phrases like "let me think," pretty much the way a person would handle them, in terms of response latency. Also, in the second half of the video, I ask the agent to adjust its response pattern because I'm going to tell it a phone number. This kind of "in-context" adjustment of response wait times is really useful. The LLM in the video is GTP-4.1. We've tested the prompt and single-token adherance with GPT-4.1, Gemini 2.5 Flash, Anthropic Claude Sonnet 4.5, and AWS Nova 2 Pro. Note that older models in all these families (and, in general, smaller open weights models) aren't able to reliably output these single-token tags. But the new models we're using these days are pretty amazing.

kwindla

26,935 views • 6 months ago

Introducing the new Box Agent. The Box Agent works across your entire Box file system, maintaining all your security and access controls, and is hyper tuned for working with enterprise content. This means you can now ask questions from all your enterprise content, search for files that were impossible to find before, deploy an agent on specific tasks on subsets of documents, analyze complex data sets, and generate or edit documents and spreadsheets via the agent. You can have the Box Agent search across your Box account to prepare for a sales meeting, analyze customer sentiment reports, process a large set of contracts for legal risk, provide insights into product development, leverage existing knowledge to answer RFPs, and thousands of other use-cases. 90% of enterprise data is unstructured data. This means most enterprise knowledge is sitting in inside of research reports, marketing assets, presentations, roadmap files, contracts, HR documents, and more. This is the critical context that agents need to be able to answer questions about a business, automate workflows, or serve up to other agents. We’ve been grinding on this for a quite a bit, and due to recent AI model advancements we’re now ready to release it to customers. Previous model generations had a difficult time knowing when to give up or keep going on a search, when to browse for files vs. use queries, how to rank files appropriately to know which version of content to use, how to handle large amounts of context to comb through, and more. Due to recent breakthroughs from models like GPT-5.4, Opus 4.6, and Gemini 3, we’ve seen major gains in tool calling, code execution, advanced reasoning, and more. Combined with an agent harness tuned to Box context, now it’s finally possible to have an agent that can work across your file system on long running tasks and actually deliver high quality results. Best of all, because the Box Agent works with any leading AI model, you’ll quickly get the gains coming out of the major labs as major new models are released. Further, openness at Box is key, so you’ll be able to call up the Box Agent from Box’s APIs and MCP server, so you can interact with Box intelligently from any other AI system. We know work happens everywhere, and we want to ensure you can access to the content you need from those places. The new Box Agent is available starting today, rolling out now for Enterprise Plus and Enterprise Advanced customers.
0:58

Introducing the new Box Agent. The Box Agent works across your entire Box file system, maintaining all your security and access controls, and is hyper tuned for working with enterprise content. This means you can now ask questions from all your enterprise content, search for files that were impossible to find before, deploy an agent on specific tasks on subsets of documents, analyze complex data sets, and generate or edit documents and spreadsheets via the agent. You can have the Box Agent search across your Box account to prepare for a sales meeting, analyze customer sentiment reports, process a large set of contracts for legal risk, provide insights into product development, leverage existing knowledge to answer RFPs, and thousands of other use-cases. 90% of enterprise data is unstructured data. This means most enterprise knowledge is sitting in inside of research reports, marketing assets, presentations, roadmap files, contracts, HR documents, and more. This is the critical context that agents need to be able to answer questions about a business, automate workflows, or serve up to other agents. We’ve been grinding on this for a quite a bit, and due to recent AI model advancements we’re now ready to release it to customers. Previous model generations had a difficult time knowing when to give up or keep going on a search, when to browse for files vs. use queries, how to rank files appropriately to know which version of content to use, how to handle large amounts of context to comb through, and more. Due to recent breakthroughs from models like GPT-5.4, Opus 4.6, and Gemini 3, we’ve seen major gains in tool calling, code execution, advanced reasoning, and more. Combined with an agent harness tuned to Box context, now it’s finally possible to have an agent that can work across your file system on long running tasks and actually deliver high quality results. Best of all, because the Box Agent works with any leading AI model, you’ll quickly get the gains coming out of the major labs as major new models are released. Further, openness at Box is key, so you’ll be able to call up the Box Agent from Box’s APIs and MCP server, so you can interact with Box intelligently from any other AI system. We know work happens everywhere, and we want to ensure you can access to the content you need from those places. The new Box Agent is available starting today, rolling out now for Enterprise Plus and Enterprise Advanced customers.

Aaron Levie

44,515 views • 5 months ago

The most downloaded AI on earth is now Chinese. Alibaba just gave away a model that matches Claude's flagship, and it literally runs on a $700 used graphics card. The Qwen models crossed 3 BILLION downloads in six months. Hugging Face counted 418 million downloads for Google this year, and 227 million for Meta. Alibaba cleared more than four times both of them combined. Then today it released Qwen3.8-27B under an Apache 2.0 license. The model has 27 billion parameters, native vision, and a 262,000 token context window. Developers are running it locally on 17 gigabytes of memory, on used cards that cost a few hundred dollars. Alibaba's own benchmark table claims it beats Opus 4.6 Max on computer use by 84.3 to 72.7, on mobile use by 81.9 to 62, and on visual math by 94.6 to 65.5. Those numbers come from the vendor and nobody has independently verified them yet, so treat them as a claim. But the generation over generation jumps are harder to wave away: On DeepSWE the score went from 13.3 to 42.2. On software engineering it went from 49.3 to 79.0. That happened in ONE release cycle. And Apache 2.0 means anyone can download the weights, modify them, build products on them, sell those products, and never pay or ask permission. It cannot be revoked. Once the file is on your drive it is yours permanently. 3 billion downloads means those files already sit on machines in every country on Earth. Alibaba could delete everything tomorrow and it would change nothing. Washington spent 4 years building an export control regime around chips, model weights, and entity lists. Every piece of it assumes a chokepoint exists somewhere. A fab, a shipment, a company that can be told no. But there is no chokepoint for a file that has already been copied three billion times. And the copying compounds. Hugging Face counted 151,448 models built on top of Qwen, which is 2.6x Meta's entire footprint and 4.7x the number of Llama repositories. New ones appear at roughly 200 a day. The report says Qwen has become "part of the default workflow for developers deciding what models to fine-tune and deploy." Alibaba is also pushing Qwen through its cloud into Southeast Asia and Africa, markets where American labs have almost no presence, and where a very large share of the next generation of developers will learn to build. Meta and Nvidia have both rushed out new open models in recent weeks. That is what a response looks like when you feel the floor move. And to be clear, these are download and derivative numbers, not usage numbers. ChatGPT and Claude cannot be downloaded at all, so they do not appear in this comparison. What the figures measure is what developers choose to build on top of, which is a different question from what consumers type into a box. That is also why it matters MORE. Consumer habits change in an afternoon. Infrastructure choices last a decade, because everything built on top has to be rewritten to undo them. The American labs are valued on an assumption that frontier intelligence stays scarce, expensive, and rented by the token. Alibaba just made a version of it free, permanent, and small enough to run on hardware people already own. You will not get an announcement when the software you use every day starts running on a Chinese model underneath. Go and count how many of the tools you rely on could be rebuilt on free weights this year.
0:39

The most downloaded AI on earth is now Chinese. Alibaba just gave away a model that matches Claude's flagship, and it literally runs on a $700 used graphics card. The Qwen models crossed 3 BILLION downloads in six months. Hugging Face counted 418 million downloads for Google this year, and 227 million for Meta. Alibaba cleared more than four times both of them combined. Then today it released Qwen3.8-27B under an Apache 2.0 license. The model has 27 billion parameters, native vision, and a 262,000 token context window. Developers are running it locally on 17 gigabytes of memory, on used cards that cost a few hundred dollars. Alibaba's own benchmark table claims it beats Opus 4.6 Max on computer use by 84.3 to 72.7, on mobile use by 81.9 to 62, and on visual math by 94.6 to 65.5. Those numbers come from the vendor and nobody has independently verified them yet, so treat them as a claim. But the generation over generation jumps are harder to wave away: On DeepSWE the score went from 13.3 to 42.2. On software engineering it went from 49.3 to 79.0. That happened in ONE release cycle. And Apache 2.0 means anyone can download the weights, modify them, build products on them, sell those products, and never pay or ask permission. It cannot be revoked. Once the file is on your drive it is yours permanently. 3 billion downloads means those files already sit on machines in every country on Earth. Alibaba could delete everything tomorrow and it would change nothing. Washington spent 4 years building an export control regime around chips, model weights, and entity lists. Every piece of it assumes a chokepoint exists somewhere. A fab, a shipment, a company that can be told no. But there is no chokepoint for a file that has already been copied three billion times. And the copying compounds. Hugging Face counted 151,448 models built on top of Qwen, which is 2.6x Meta's entire footprint and 4.7x the number of Llama repositories. New ones appear at roughly 200 a day. The report says Qwen has become "part of the default workflow for developers deciding what models to fine-tune and deploy." Alibaba is also pushing Qwen through its cloud into Southeast Asia and Africa, markets where American labs have almost no presence, and where a very large share of the next generation of developers will learn to build. Meta and Nvidia have both rushed out new open models in recent weeks. That is what a response looks like when you feel the floor move. And to be clear, these are download and derivative numbers, not usage numbers. ChatGPT and Claude cannot be downloaded at all, so they do not appear in this comparison. What the figures measure is what developers choose to build on top of, which is a different question from what consumers type into a box. That is also why it matters MORE. Consumer habits change in an afternoon. Infrastructure choices last a decade, because everything built on top has to be rewritten to undo them. The American labs are valued on an assumption that frontier intelligence stays scarce, expensive, and rented by the token. Alibaba just made a version of it free, permanent, and small enough to run on hardware people already own. You will not get an announcement when the software you use every day starts running on a Chinese model underneath. Go and count how many of the tools you rely on could be rebuilt on free weights this year.

Ricardo

80,611 views • 13 days ago

Mark Zuckerberg is explaining one of the most misunderstood dynamics in AI and it has direct investment implications (Save this). The concept he's describing is model distillation, and it's one of the most important techniques to emerge in AI over the past year. Here's how it works. You train a massive, enormously expensive model, in Meta's case, Llama 4 Behemoth, a 2 trillion parameter teacher model and then you use that model to teach a much smaller, cheaper model. The smaller model inherits roughly 90 to 95% of the intelligence of the giant while running at 10% of the cost and on a fraction of the compute. Meta already did this with the Llama 4 family and Behemoth serves as the teacher. Llama 4 Scout and Maverick, the publicly released open-source models were distilled from it. Scout runs on a single H100 GPU with a 10 million token context window and outperforms models that cost far more to operate. Maverick, at 17 billion active parameters, rivals DeepSeek V3 in coding at half the parameter count and beats GPT-4o on multimodal benchmarks. Both are completely free for commercial use. What Zuckerberg is pointing at is a structural shift in how AI gets deployed in the real world. Companies aren't taking a frontier model off the shelf and running it as-is but rather taking open-source models, fine-tuning them on their own proprietary data, distilling them into even smaller custom models tailored to their specific use case, and running them on infrastructure they control at a fraction of the cost of a closed frontier API. The investment implication of this is significant and runs in two directions. For Meta specifically, this is a strategic masterstroke. Every company that builds on Llama, fine-tunes it, distills it, or deploys it through their infrastructure is pulling into Meta's orbit while Meta builds the most powerful open teacher model. The ecosystem of companies using it grows and that ecosystem generates commercial activity across Meta's platforms and data services. Meta's AI research benefits from billions of real world deployment signals and it's a flywheel that closed model providers cannot replicate because their strategy requires charging per token, which is now a 65x cost disadvantage against the open-source alternative. For the broader market, distillation changes the economics of inference in a way that has barely been priced in. As intelligence becomes extractable into smaller and cheaper models, the absolute demand for compute doesn't decline but rather it explodes, because now the number of applications that are economically viable expands by orders of magnitude. Every task that was previously too expensive to automate at $3.25 per call becomes viable at $0.05 that means more total token usage, more total GPU utilization, and more demand for the infrastructure companies, the Nebiuses, the GE Vernovas, the Constellation Energies that supply the underlying compute and power.
1:05

Mark Zuckerberg is explaining one of the most misunderstood dynamics in AI and it has direct investment implications (Save this). The concept he's describing is model distillation, and it's one of the most important techniques to emerge in AI over the past year. Here's how it works. You train a massive, enormously expensive model, in Meta's case, Llama 4 Behemoth, a 2 trillion parameter teacher model and then you use that model to teach a much smaller, cheaper model. The smaller model inherits roughly 90 to 95% of the intelligence of the giant while running at 10% of the cost and on a fraction of the compute. Meta already did this with the Llama 4 family and Behemoth serves as the teacher. Llama 4 Scout and Maverick, the publicly released open-source models were distilled from it. Scout runs on a single H100 GPU with a 10 million token context window and outperforms models that cost far more to operate. Maverick, at 17 billion active parameters, rivals DeepSeek V3 in coding at half the parameter count and beats GPT-4o on multimodal benchmarks. Both are completely free for commercial use. What Zuckerberg is pointing at is a structural shift in how AI gets deployed in the real world. Companies aren't taking a frontier model off the shelf and running it as-is but rather taking open-source models, fine-tuning them on their own proprietary data, distilling them into even smaller custom models tailored to their specific use case, and running them on infrastructure they control at a fraction of the cost of a closed frontier API. The investment implication of this is significant and runs in two directions. For Meta specifically, this is a strategic masterstroke. Every company that builds on Llama, fine-tunes it, distills it, or deploys it through their infrastructure is pulling into Meta's orbit while Meta builds the most powerful open teacher model. The ecosystem of companies using it grows and that ecosystem generates commercial activity across Meta's platforms and data services. Meta's AI research benefits from billions of real world deployment signals and it's a flywheel that closed model providers cannot replicate because their strategy requires charging per token, which is now a 65x cost disadvantage against the open-source alternative. For the broader market, distillation changes the economics of inference in a way that has barely been priced in. As intelligence becomes extractable into smaller and cheaper models, the absolute demand for compute doesn't decline but rather it explodes, because now the number of applications that are economically viable expands by orders of magnitude. Every task that was previously too expensive to automate at $3.25 per call becomes viable at $0.05 that means more total token usage, more total GPU utilization, and more demand for the infrastructure companies, the Nebiuses, the GE Vernovas, the Constellation Energies that supply the underlying compute and power.

Milk Road AI

27,908 views • 1 month ago

computer use agents are slow because they think too much they make round trips to an LLM. every single time. even for a command you've said a hundred times before. that's why it always feels slow. my mac now remembers. the first time i say something new, like "search youtube for lofi beats", a big model figures it out once, does it, and saves the recipe as a json spec: {"op":"act","do":"open","target":"Safari"} {"op":"act","do":"type","text":"{query}"} every time after that, a tiny model on my laptop recognizes the command, fills in the parameters, and replays the recipe with CUA say it once, it's slow. say it again, it's instant. the fine tuned tiny model is 350 million parameters. its only job is recognizing which of my commands i just said. and when the big model does plan, it streams actions line by line. each line runs the moment it's finished, so the first action is executing while the model is still writing the second. everything it learns lives in one plain text file on my disk. i've been playing around a lot with Cua by Francesco and peekaboo by Peter Steinberger 🦞 . curious whether a standard computer use spec makes sense, one that streaming models can drive to make computer use actually fast.
3:11

computer use agents are slow because they think too much they make round trips to an LLM. every single time. even for a command you've said a hundred times before. that's why it always feels slow. my mac now remembers. the first time i say something new, like "search youtube for lofi beats", a big model figures it out once, does it, and saves the recipe as a json spec: {"op":"act","do":"open","target":"Safari"} {"op":"act","do":"type","text":"{query}"} every time after that, a tiny model on my laptop recognizes the command, fills in the parameters, and replays the recipe with CUA say it once, it's slow. say it again, it's instant. the fine tuned tiny model is 350 million parameters. its only job is recognizing which of my commands i just said. and when the big model does plan, it streams actions line by line. each line runs the moment it's finished, so the first action is executing while the model is still writing the second. everything it learns lives in one plain text file on my disk. i've been playing around a lot with Cua by Francesco and peekaboo by Peter Steinberger 🦞 . curious whether a standard computer use spec makes sense, one that streaming models can drive to make computer use actually fast.

Milind S

14,212 views • 1 month ago

Most video-action robot models are a content-creation video generator with an action module attached. LingBot-VA 2.0 from Robbyant, a video-action foundation model, throws that starting point out and trains the whole stack natively for control. And it runs closed-loop at a peak 225 Hz. It's so important because A robot cannot move responsively when its controller pauses to imagine the next few frames. LingBot-VA 2.0 predicts during execution, then corrects using each real observation. And it carries only about 13B video parameters while activating roughly 1.9B per token. Bigger robot models usually mean slower reactions, creating a direct conflict between intelligence and control. LingBot-VA 2.0 is trained from scratch for robot control rather than adapted from a video generator built for content creation. Robbyant, an embodied AI company under Ant Group, built it to learn how scenes change under actions, predict what should happen next, and turn those predictions into real-time robot movements. Most video-action systems inherit a tokenizer and video backbone trained mainly to reproduce visual appearance. LingBot-VA 2.0 rebuilds both parts around physical control. Its semantic visual-action tokenizer maps observations toward features from a frozen vision foundation model and learns compact latent actions from frame-to-frame changes using self-supervised inverse and forward dynamics. Unlabeled web video can therefore carry action-relevant training signals without robot action labels. The policy is causal from the start, so every prediction can use only past observations. Its sparse Mixture-of-Experts video backbone has about 13B total parameters, while about 1.9B are active per token, keeping the compute lower during each step. A high-level vision-language planner breaks long tasks into smaller instructions, while the low-level video-action policy handles continuous movement. Foresight Reasoning predicts future visual states while the robot is already acting, then replaces imagined states with every new real observation. Combined with few-step distillation and systems acceleration, the paper reports a peak asynchronous execution frequency of 225 Hz. The model adapts from 10–15 demonstrations, transfers across robot embodiments, and handles some new tasks zero-shot. In the paper’s own evaluations, it reaches 93.6 average on RoboTwin 2.0 and reports stronger real-world results than LingBot-VA and π0.5 across the tested tasks. 🧵 1.
1:17

Most video-action robot models are a content-creation video generator with an action module attached. LingBot-VA 2.0 from Robbyant, a video-action foundation model, throws that starting point out and trains the whole stack natively for control. And it runs closed-loop at a peak 225 Hz. It's so important because A robot cannot move responsively when its controller pauses to imagine the next few frames. LingBot-VA 2.0 predicts during execution, then corrects using each real observation. And it carries only about 13B video parameters while activating roughly 1.9B per token. Bigger robot models usually mean slower reactions, creating a direct conflict between intelligence and control. LingBot-VA 2.0 is trained from scratch for robot control rather than adapted from a video generator built for content creation. Robbyant, an embodied AI company under Ant Group, built it to learn how scenes change under actions, predict what should happen next, and turn those predictions into real-time robot movements. Most video-action systems inherit a tokenizer and video backbone trained mainly to reproduce visual appearance. LingBot-VA 2.0 rebuilds both parts around physical control. Its semantic visual-action tokenizer maps observations toward features from a frozen vision foundation model and learns compact latent actions from frame-to-frame changes using self-supervised inverse and forward dynamics. Unlabeled web video can therefore carry action-relevant training signals without robot action labels. The policy is causal from the start, so every prediction can use only past observations. Its sparse Mixture-of-Experts video backbone has about 13B total parameters, while about 1.9B are active per token, keeping the compute lower during each step. A high-level vision-language planner breaks long tasks into smaller instructions, while the low-level video-action policy handles continuous movement. Foresight Reasoning predicts future visual states while the robot is already acting, then replaces imagined states with every new real observation. Combined with few-step distillation and systems acceleration, the paper reports a peak asynchronous execution frequency of 225 Hz. The model adapts from 10–15 demonstrations, transfers across robot embodiments, and handles some new tasks zero-shot. In the paper’s own evaluations, it reaches 93.6 average on RoboTwin 2.0 and reports stronger real-world results than LingBot-VA and π0.5 across the tested tasks. 🧵 1.

Rohan Paul

11,253 views • 1 month ago

China just released an open source AI model that matches the best closed models from OpenAI and Anthropic. Gavin Baker explained exactly how they did it and the answer should concern every American AI lab. The model is called GLM 5.2. It was built by Z. AI. You get 744 billion parameters, 1 million token context window and its MIT license, meaning anyone can download it, fork it, build a company on it, with no restrictions and no Dario. It scored 51 points on the artificial analysis intelligence index. The highest score any open weight model has ever achieved. It beat GPT 5.5 on the frontier software engineering benchmark. It trails Claude Opus 4.8 by less than one percentage point. And it costs 85% less to run than GPT 5.5 for comparable performance. Gavin Baker said on the All-In podcast that this model has challenged some of his beliefs. Then he explained how China built it. The method is called distillation. Just think of tens of thousands of phones and computers running simultaneously, all hitting the frontier model APIs through masked accounts, asking specific questions, and harvesting what happens inside the model when it answers. Every reasoning step, every token. The entire thinking process gets recorded and fed back into the Chinese model during training. It is a cheat sheet. It is the answer key to the exam. And here is the part that should worry everyone. Sacks said it plainly. China was already nine months behind American models. But now that GLM 5.2 is good enough to run its own reinforcement learning, it can improve itself without needing to distill from American models anymore. The cheat sheet let them get close enough to start writing their own answers. Sacks said we are six months behind on the model and 24 months behind on silicon and they are only a few months behind in total. The Z. AI founder told Elon Musk directly that open weight fable-level capability will be here before Q1 2027. Every restriction Anthropic lobbied for, every self-imposed safety guardrail, every month of delay in releasing American frontier models accelerated this. The Chinese labs were not under those restrictions. They were not going to wait. The composable model future Gavin described, where every enterprise runs a frontier model alongside their own fine-tuned open weight model, is coming regardless of what American labs do next. The question is just whether the open weight half of that stack is American or Chinese. Right now it is Chinese. WATCH THE FULL PODCAST ON The All-In Podcast
3:21

China just released an open source AI model that matches the best closed models from OpenAI and Anthropic. Gavin Baker explained exactly how they did it and the answer should concern every American AI lab. The model is called GLM 5.2. It was built by Z. AI. You get 744 billion parameters, 1 million token context window and its MIT license, meaning anyone can download it, fork it, build a company on it, with no restrictions and no Dario. It scored 51 points on the artificial analysis intelligence index. The highest score any open weight model has ever achieved. It beat GPT 5.5 on the frontier software engineering benchmark. It trails Claude Opus 4.8 by less than one percentage point. And it costs 85% less to run than GPT 5.5 for comparable performance. Gavin Baker said on the All-In podcast that this model has challenged some of his beliefs. Then he explained how China built it. The method is called distillation. Just think of tens of thousands of phones and computers running simultaneously, all hitting the frontier model APIs through masked accounts, asking specific questions, and harvesting what happens inside the model when it answers. Every reasoning step, every token. The entire thinking process gets recorded and fed back into the Chinese model during training. It is a cheat sheet. It is the answer key to the exam. And here is the part that should worry everyone. Sacks said it plainly. China was already nine months behind American models. But now that GLM 5.2 is good enough to run its own reinforcement learning, it can improve itself without needing to distill from American models anymore. The cheat sheet let them get close enough to start writing their own answers. Sacks said we are six months behind on the model and 24 months behind on silicon and they are only a few months behind in total. The Z. AI founder told Elon Musk directly that open weight fable-level capability will be here before Q1 2027. Every restriction Anthropic lobbied for, every self-imposed safety guardrail, every month of delay in releasing American frontier models accelerated this. The Chinese labs were not under those restrictions. They were not going to wait. The composable model future Gavin described, where every enterprise runs a frontier model alongside their own fine-tuned open weight model, is coming regardless of what American labs do next. The question is just whether the open weight half of that stack is American or Chinese. Right now it is Chinese. WATCH THE FULL PODCAST ON The All-In Podcast

Ihtesham Ali

86,452 views • 2 months ago

HERMES AGENT SUPPORTS 300+ MODELS. PICKING THE RIGHT ONE PER TASK IS THE DIFFERENCE BETWEEN $5/MONTH AND $50. STARTING OUT: Claude Sonnet 4.6. official recommendation from Nous Research. "the model this project was built and tested with." strong reasoning. reliable tool calling. mid-range pricing. PREMIUM TIER: Claude Opus 4.8. best coding benchmarks available. self-correcting reasoning. catches its own mistakes. 1M context. use for demanding tasks where quality matters. GPT-5.5. #1 Chatbot Arena. #1 GPQA Diamond reasoning (94.1%). #1 creative writing. 2M context. handles entire codebases in one pass. Grok 4.30. the only frontier model with live X firehose access. real-time social data, breaking news, market sentiment. connects via Grok OAuth. no separate API key. Grok-Composer-2.5-Fast (v0.17.0). Cursor's coding model. 200K context. available through your Grok subscription via OAuth. no extra cost if you already pay for Grok. MID-RANGE TIER: Claude Sonnet 4.6. best balance of quality and cost for daily use. strongest prose and tool calling in this tier. Gemini 2.5 Pro. Google Search grounding built in. cites sources. verifies claims. pulls current data. 2M context. best for research-heavy workflows. GPT-4.1. reliable tool calling. solid general reasoning. good middle ground when you need OpenAI compatibility. BUDGET TIER: Claude Haiku 4.5. fastest Anthropic model. cheapest paid Claude option. strong at classification, routing, simple queries. use for auxiliary tasks: compression, vision, web extraction, approval scoring. DeepSeek V4. best cost-to-quality ratio in the market. 90% cache discount on repeated context. use for sub-agents and bulk parallel work. DeepSeek V4 Flash. cheapest paid model worth using. 1M context. MIT license. self-hostable. use for cron jobs, monitoring, routine searches. MiniMax M3. Nous Research and MiniMax collaborating on optimization. 1M context via lightning attention. 59% SWE-Bench Pro. beats several premium models on coding. one of the most-used models inside Hermes. FREE / LOCAL: Qwen 3.5 27B via Ollama. 16GB VRAM. reliable tool calling. best free local model for Hermes as of mid-2026. Qwen 3 8B. 8GB VRAM. fits a $7 VPS. handles routine tasks at zero API cost. Llama 4 Maverick. best open-weight tool calling. 1M context. needs more VRAM but strongest local option. HOW TO ASSIGN MODELS: main model: Desktop app / Dashboard → Models → switch sub-agent model: set in Desktop app, Dashboard, or config.yaml: delegation: model: "deepseek/deepseek-v4" auxiliary models (compression, vision, web extract): Desktop app / Dashboard → Models → Auxiliary Haiku 4.5 or Gemini Flash work well here. saves significantly when your main model is premium. per-profile: each Hermes profile gets its own model. Scout on DeepSeek. Analyst on Sonnet. Briefer on budget model. Coder on Opus. per-cron-job: pin a specific model to any cron job. morning brief on Haiku. deep research on Sonnet. monitoring on DeepSeek Flash. each job uses only the model it needs. per-session: /model deepseek/deepseek-v4-flash hot-swap mid-conversation. no restart needed. FALLBACK CHAINS: if your primary model is unavailable, Hermes automatically switches to the next provider. rate limit or server error = next model in the chain. no failed runs. no manual intervention. set in Desktop app, Dashboard, or config.yaml: fallback_providers: - openrouter - nous - codex PROVIDER PATHS: OPENROUTER: 300+ models under one API key. pay per token. most flexible. NOUS PORTAL: 300+ models + Tool Gateway (web search, image gen, TTS, browser). one OAuth. one subscription. 10% off token-billed providers. CHATGPT SUB: GPT-5.5 + Grok via OAuth. included tokens with $20 subscription. OLLAMA: free. local. private. zero API cost. your hardware only. mix providers across profiles and tasks. Scout on OpenRouter. Analyst on Nous Portal. Coder on ChatGPT sub. Monitor on Ollama. THE RULE: premium for work that needs deep reasoning. mid-range for daily driver tasks. budget for volume and background work. free for monitoring and routine jobs. pricing changes fast. check openrouter ai for current rates before committing. Which is your favourite model and for what task? full 15 levels breakdown in the article 👇
0:51

HERMES AGENT SUPPORTS 300+ MODELS. PICKING THE RIGHT ONE PER TASK IS THE DIFFERENCE BETWEEN $5/MONTH AND $50. STARTING OUT: Claude Sonnet 4.6. official recommendation from Nous Research. "the model this project was built and tested with." strong reasoning. reliable tool calling. mid-range pricing. PREMIUM TIER: Claude Opus 4.8. best coding benchmarks available. self-correcting reasoning. catches its own mistakes. 1M context. use for demanding tasks where quality matters. GPT-5.5. #1 Chatbot Arena. #1 GPQA Diamond reasoning (94.1%). #1 creative writing. 2M context. handles entire codebases in one pass. Grok 4.30. the only frontier model with live X firehose access. real-time social data, breaking news, market sentiment. connects via Grok OAuth. no separate API key. Grok-Composer-2.5-Fast (v0.17.0). Cursor's coding model. 200K context. available through your Grok subscription via OAuth. no extra cost if you already pay for Grok. MID-RANGE TIER: Claude Sonnet 4.6. best balance of quality and cost for daily use. strongest prose and tool calling in this tier. Gemini 2.5 Pro. Google Search grounding built in. cites sources. verifies claims. pulls current data. 2M context. best for research-heavy workflows. GPT-4.1. reliable tool calling. solid general reasoning. good middle ground when you need OpenAI compatibility. BUDGET TIER: Claude Haiku 4.5. fastest Anthropic model. cheapest paid Claude option. strong at classification, routing, simple queries. use for auxiliary tasks: compression, vision, web extraction, approval scoring. DeepSeek V4. best cost-to-quality ratio in the market. 90% cache discount on repeated context. use for sub-agents and bulk parallel work. DeepSeek V4 Flash. cheapest paid model worth using. 1M context. MIT license. self-hostable. use for cron jobs, monitoring, routine searches. MiniMax M3. Nous Research and MiniMax collaborating on optimization. 1M context via lightning attention. 59% SWE-Bench Pro. beats several premium models on coding. one of the most-used models inside Hermes. FREE / LOCAL: Qwen 3.5 27B via Ollama. 16GB VRAM. reliable tool calling. best free local model for Hermes as of mid-2026. Qwen 3 8B. 8GB VRAM. fits a $7 VPS. handles routine tasks at zero API cost. Llama 4 Maverick. best open-weight tool calling. 1M context. needs more VRAM but strongest local option. HOW TO ASSIGN MODELS: main model: Desktop app / Dashboard → Models → switch sub-agent model: set in Desktop app, Dashboard, or config.yaml: delegation: model: "deepseek/deepseek-v4" auxiliary models (compression, vision, web extract): Desktop app / Dashboard → Models → Auxiliary Haiku 4.5 or Gemini Flash work well here. saves significantly when your main model is premium. per-profile: each Hermes profile gets its own model. Scout on DeepSeek. Analyst on Sonnet. Briefer on budget model. Coder on Opus. per-cron-job: pin a specific model to any cron job. morning brief on Haiku. deep research on Sonnet. monitoring on DeepSeek Flash. each job uses only the model it needs. per-session: /model deepseek/deepseek-v4-flash hot-swap mid-conversation. no restart needed. FALLBACK CHAINS: if your primary model is unavailable, Hermes automatically switches to the next provider. rate limit or server error = next model in the chain. no failed runs. no manual intervention. set in Desktop app, Dashboard, or config.yaml: fallback_providers: - openrouter - nous - codex PROVIDER PATHS: OPENROUTER: 300+ models under one API key. pay per token. most flexible. NOUS PORTAL: 300+ models + Tool Gateway (web search, image gen, TTS, browser). one OAuth. one subscription. 10% off token-billed providers. CHATGPT SUB: GPT-5.5 + Grok via OAuth. included tokens with $20 subscription. OLLAMA: free. local. private. zero API cost. your hardware only. mix providers across profiles and tasks. Scout on OpenRouter. Analyst on Nous Portal. Coder on ChatGPT sub. Monitor on Ollama. THE RULE: premium for work that needs deep reasoning. mid-range for daily driver tasks. budget for volume and background work. free for monitoring and routine jobs. pricing changes fast. check openrouter ai for current rates before committing. Which is your favourite model and for what task? full 15 levels breakdown in the article 👇

YanXbt

17,138 views • 2 months ago

New model: your robot can now pack your suitcase 🧳 Xiaomi has released a new robot foundation model. Called Xiaomi-Robotics-1, it is designed to have a robot pick things up and move them around. But first, DEFINITIONS: - Mixture-of-Transformers (MoT): An architecture where separate transformer "experts" (e.g., one for vision-language, one for actions) share a single attention stream, so each modality gets specialized parameters without losing joint reasoning. - Vision-language model (VLM): A model that jointly understands images and text. - Diffusion transformer: A transformer trained to turn noise into structured outputs by iterative denoising, here generating robot actions rather than images. - Action chunks: Short sequences of future actions (e.g., the next ~50 motor commands) predicted in one shot instead of one step at a time. - Flow matching: A faster version of diffusion. The model learns a straight-line velocity field from noise to the target action, so it needs only a few integration steps instead of many denoising ones. Its peculiarity comes from its two stage training: 1. 100,000 hours of video shot through a UMI rig: a handheld 3D-printed gripper with a camera, worn by humans doing ordinary tasks in homes, shops, factories and offices. 2. Adapt to actual robot bodies with ~10,000 hours of real-robot data. It replaces the standard approach of teleoperating a real robot for every hour of training data. Its architecture is a Mixture-of-Transformers pairing a pre-trained Qwen3-VL vision-language model with a diffusion transformer that emits action chunks via flow matching, released in 2.6B, 5.1B and 10.5B parameter variants. However, if you read the entire paper ("Scaling VLA Models with over 100K Hours"), you realize that all of the scaling experiments on 20k hours. Therefore the headline "out-of-the-box success climbing 26% → 75% as pre-training data grows" tops out at 100% of 20k hours! What the full corpus does to that curve is never shown -> and this where things would become interesting! Xiaomi's own conclusion is that model size has stopped mattering and data is the binding constraint. The performance gap among different model sizes are less pronounced than those observed across different data scales. This result suggests that model capacity at the billions-parameter scale may already be sufficient to capture the current dataset's distribution. Which further asks the same question: why not use the 100k video hours? Anyway, I would definitely love to have a couple robots at home that can cooperate to pack my suitcase with items relevant to my next destination:
0:58

New model: your robot can now pack your suitcase 🧳 Xiaomi has released a new robot foundation model. Called Xiaomi-Robotics-1, it is designed to have a robot pick things up and move them around. But first, DEFINITIONS: - Mixture-of-Transformers (MoT): An architecture where separate transformer "experts" (e.g., one for vision-language, one for actions) share a single attention stream, so each modality gets specialized parameters without losing joint reasoning. - Vision-language model (VLM): A model that jointly understands images and text. - Diffusion transformer: A transformer trained to turn noise into structured outputs by iterative denoising, here generating robot actions rather than images. - Action chunks: Short sequences of future actions (e.g., the next ~50 motor commands) predicted in one shot instead of one step at a time. - Flow matching: A faster version of diffusion. The model learns a straight-line velocity field from noise to the target action, so it needs only a few integration steps instead of many denoising ones. Its peculiarity comes from its two stage training: 1. 100,000 hours of video shot through a UMI rig: a handheld 3D-printed gripper with a camera, worn by humans doing ordinary tasks in homes, shops, factories and offices. 2. Adapt to actual robot bodies with ~10,000 hours of real-robot data. It replaces the standard approach of teleoperating a real robot for every hour of training data. Its architecture is a Mixture-of-Transformers pairing a pre-trained Qwen3-VL vision-language model with a diffusion transformer that emits action chunks via flow matching, released in 2.6B, 5.1B and 10.5B parameter variants. However, if you read the entire paper ("Scaling VLA Models with over 100K Hours"), you realize that all of the scaling experiments on 20k hours. Therefore the headline "out-of-the-box success climbing 26% → 75% as pre-training data grows" tops out at 100% of 20k hours! What the full corpus does to that curve is never shown -> and this where things would become interesting! Xiaomi's own conclusion is that model size has stopped mattering and data is the binding constraint. The performance gap among different model sizes are less pronounced than those observed across different data scales. This result suggests that model capacity at the billions-parameter scale may already be sufficient to capture the current dataset's distribution. Which further asks the same question: why not use the 100k video hours? Anyway, I would definitely love to have a couple robots at home that can cooperate to pack my suitcase with items relevant to my next destination:

Léo

15,662 views • 25 days ago

Web scraping will never be the same. (100% open-source visual search at scale) PixelRAG is a retrieval system that skips HTML parsing completely. Instead of scraping a page into text and embedding chunks, it screenshots the page and retrieves the image. A vision-language model reads the answer straight off the pixels. Why that matters: parsing is where web RAG quietly loses information. - A single HTML-to-text parser can drop 40%+ of a page. - Tables, charts, and layout get flattened or thrown out. - Swapping parsers alone can move accuracy ~10 points on the same docs. PixelRAG indexes the page a person actually sees. The team built a visual index of all of Wikipedia, 30M+ screenshots, and it still beats the strongest text RAG baseline by 18.1% on text-only QA. The repo also ships a Claude Code plugin that gives Claude eyes. It lets Claude screenshot any URL and read the rendered page instead of scraping the DOM. So you can hand it a live page, an arXiv paper, or your local site and ask what it actually looks like. One setup script. No MCP server, no backend. How the pipeline works: - Renders each document (web, PDF, image) to image tiles. - Embeds them with Qwen3-VL-Embedding, LoRA fine-tuned on screenshots. - Builds a FAISS index and serves a search API. A stronger reader model lifts accuracy with no re-indexing, since the index is just pixels. Everything is open-source under Apache-2.0. GitHub repo: Talking about RAG, I recently wrote an article on a new approach that makes retrieval much more efficient by cutting corpus size by 40x, reducing tokens per query by 3x, and improving vector search relevance by 2.3x. The article is quoted below.
0:55

Web scraping will never be the same. (100% open-source visual search at scale) PixelRAG is a retrieval system that skips HTML parsing completely. Instead of scraping a page into text and embedding chunks, it screenshots the page and retrieves the image. A vision-language model reads the answer straight off the pixels. Why that matters: parsing is where web RAG quietly loses information. - A single HTML-to-text parser can drop 40%+ of a page. - Tables, charts, and layout get flattened or thrown out. - Swapping parsers alone can move accuracy ~10 points on the same docs. PixelRAG indexes the page a person actually sees. The team built a visual index of all of Wikipedia, 30M+ screenshots, and it still beats the strongest text RAG baseline by 18.1% on text-only QA. The repo also ships a Claude Code plugin that gives Claude eyes. It lets Claude screenshot any URL and read the rendered page instead of scraping the DOM. So you can hand it a live page, an arXiv paper, or your local site and ask what it actually looks like. One setup script. No MCP server, no backend. How the pipeline works: - Renders each document (web, PDF, image) to image tiles. - Embeds them with Qwen3-VL-Embedding, LoRA fine-tuned on screenshots. - Builds a FAISS index and serves a search API. A stronger reader model lifts accuracy with no re-indexing, since the index is just pixels. Everything is open-source under Apache-2.0. GitHub repo: Talking about RAG, I recently wrote an article on a new approach that makes retrieval much more efficient by cutting corpus size by 40x, reducing tokens per query by 3x, and improving vector search relevance by 2.3x. The article is quoted below.

Akshay 🚀

946,169 views • 2 months ago