Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Wait Xiaomi has also released the best 9B model in the world?! This model is based on Qwen-9B... And you can easily run it even on 8GB of memory! - MIT license - Multimodal (image + text) - Solid for coding/agentic tasks Perfect to use locally and offline... even... show more

Paul Couvert

220,761 subscribers

82,735 views • 1 day ago •via X (Twitter)

Education Science & Technology

Anya Rossi• Live Now

Private livecam show

24 Comments

Paul Couvert1 day ago

Base model weights: Quantized GGUF:

Deepakraj Solanki1 day ago

Stop making wrong claims and that 8GB RAM thing is just so stupid about you

Paul Couvert1 day ago

Fits using a Q4 version, especially if you're offloading the context in the ram

Nikolaus Kern1 day ago

Amazing. Now let's see your tests on 8gb VRAM 🙃

Antifield1 day ago

我认为并不是，Ornith-1.5-9B在编码任务上明显更强

Mac1 day ago

@grok will this work on my pc? I have rtx 3060 and 32 GB RAM

Jimmy Otis #TruthMatters #EndCronyism1 day ago

max-distilled fable.... nice.

Paul Couvert1 day ago

Everyone is distilling from everyone... and even closed labs are integrating open research.

Jimmy Otis #TruthMatters #EndCronyism1 day ago

yes. IMO anthropic has been distilled most. Meta distills least from other models. A bit like inbreeding. We need more diversity!

Mark Nisin1 day ago

This one is very good and high quality!

AI News Daily1 day ago

The 8GB angle is what makes this genuinely useful, not just impressive on a benchmark. I am curious how it holds up on longer agentic tasks once context and tool calls start competing for memory.

Fajar M Reza1 day ago

Eight-gigabyte multimodal inference makes privacy concrete, but leaves little context headroom.

Simon Bullows1 day ago

@grok how would this compare to spark 2.5 4b in terms of performance and requirements? If I have 12bg if VRAM, what tokens a second could I expect?

Alex Radulescu 🇸🇬1 day ago

Have you tested it?

Paul Couvert1 day ago

On it. So far so good.

Crio Songo1 day ago

That's really impressive. Open-source, low-resource requirement, multi-modal, great for local deployment.

NotaDEV1 day ago

can it actually hold a long agentic loop or does it fall apart after step three

Quinn’s Neural Pathways1 day ago

GPUs go burrr is optional now. A 9B MIT-licensed agentic model that runs offline on 8GB of unified memory is the actual news here.

Abhiram.N.S1 day ago

I think bonsai 2 will still be on top due to the 27B representational space .

ఏది అయితే ఏంటి1 day ago

Will it work on non Mac

Snapolino1 day ago

unfortunately the RL version is not yet released :-) only the SFT versino

Brjan | AI Builder1 day ago

how does it handle complex coding tasks compared to other models?

Enoch Bowden1 day ago

Wait... Fuck off

Laythe1 day ago

is it actually any good though

Related Videos

NEW: Alibaba just released Qwen 3.5 Small — a family of powerful multimodal models available in a range of sizes (0.8B, 2B, 4B, and 9B parameters). Perfect for on-device applications! They can even run 100% locally in your browser on WebGPU, powered by Transformers.js! 🤯

NEW: Alibaba just released Qwen 3.5 Small — a family of powerful multimodal models available in a range of sizes (0.8B, 2B, 4B, and 9B parameters). Perfect for on-device applications! They can even run 100% locally in your browser on WebGPU, powered by Transformers.js! 🤯

Xenova

102,592 views • 6 months ago

OpenAI has finally released GPT OSS 20B!! This is probably the most powerful model you can run locally on a laptop. → Only 3.6B active parameters → Adjustable "Reasoning Effort" → Works with 16GB of RAM! Quick steps to use it 100% free and offline:

OpenAI has finally released GPT OSS 20B!! This is probably the most powerful model you can run locally on a laptop. → Only 3.6B active parameters → Adjustable "Reasoning Effort" → Works with 16GB of RAM! Quick steps to use it 100% free and offline:

Paul Couvert

595,738 views • 1 year ago

Introducing MTPLX V2.12: The fastest way to run local AI now even faster and with more models for lower memory macs and the new M5 Ultra. Upgrades: - 100 TPS Qwen Flash Next sustained on long coding (M5 max). - 2x Prefill speeds Qwen Flash Next: 872 --> 1695 TPS. - Requests with images now use the compiled kernels for max speeds. New Models: - Qwen Flash Next MTPLX Optimized Quality: 8 bit model built for the M5 Ultra and M3 Ultra. - Bonsai 2 27B MTPLX Optimized Speed: Works on 16gb and 24gb macs with compiled kernels and MTP running as fast as possible. 64 TPS. - MiMo V2.6 Qwen 3.5 9B MTPLX Optimized Speed: My personal favourite. Mimos magic distilled into a 9B model easily runnable at high speeds on a 16gb mac. 107 TPS. +40 bug fixes and improvements in real world coding.

Introducing MTPLX V2.12: The fastest way to run local AI now even faster and with more models for lower memory macs and the new M5 Ultra. Upgrades: - 100 TPS Qwen Flash Next sustained on long coding (M5 max). - 2x Prefill speeds Qwen Flash Next: 872 --> 1695 TPS. - Requests with images now use the compiled kernels for max speeds. New Models: - Qwen Flash Next MTPLX Optimized Quality: 8 bit model built for the M5 Ultra and M3 Ultra. - Bonsai 2 27B MTPLX Optimized Speed: Works on 16gb and 24gb macs with compiled kernels and MTP running as fast as possible. 64 TPS. - MiMo V2.6 Qwen 3.5 9B MTPLX Optimized Speed: My personal favourite. Mimos magic distilled into a 9B model easily runnable at high speeds on a 16gb mac. 107 TPS. +40 bug fixes and improvements in real world coding.

Youssof Al Toukhi

20,005 views • 2 days ago

Today we’re taking a big step on the path toward AGI and releasing Gemini 3— our most intelligent model yet. With Gemini 3, you can bring any idea to life. It is state-of-the-art in reasoning, the best model in the world for multimodal understanding, and our best agentic and vibe coding model.

Today we’re taking a big step on the path toward AGI and releasing Gemini 3— our most intelligent model yet. With Gemini 3, you can bring any idea to life. It is state-of-the-art in reasoning, the best model in the world for multimodal understanding, and our best agentic and vibe coding model.

Google AI

493,268 views • 10 months ago

DeepSeek has just released a powerful new 8B model 🔥 This means you can run it locally and for free on a laptop with zero limitations. You just need LM Studio and 4Gb of RAM. The few quick steps below:

DeepSeek has just released a powerful new 8B model 🔥 This means you can run it locally and for free on a laptop with zero limitations. You just need LM Studio and 4Gb of RAM. The few quick steps below:

Paul Couvert

370,859 views • 1 year ago

The first truly open-source audio-video model. LTX-2 is a DiT-based foundation model with all core video generation capabilities in one unified model. Designed to run locally on consumer GPUs. - text-to-video - image-to-video - and video-to-video modes 100% open-source.

The first truly open-source audio-video model. LTX-2 is a DiT-based foundation model with all core video generation capabilities in one unified model. Designed to run locally on consumer GPUs. - text-to-video - image-to-video - and video-to-video modes 100% open-source.

Akshay 🚀

66,197 views • 8 months ago

I seriously cannot believe this is a 0.6B LLM! 🤯 Qwen just released Qwen3, a series of hybrid reasoning models that allow you to control how much "thinking" the model does for a given task. They can even run locally in your browser on WebGPU with 🤗 Transformers.js!

I seriously cannot believe this is a 0.6B LLM! 🤯 Qwen just released Qwen3, a series of hybrid reasoning models that allow you to control how much "thinking" the model does for a given task. They can even run locally in your browser on WebGPU with 🤗 Transformers.js!

Xenova

74,954 views • 1 year ago

i gave a 9B model an agentic OS and told it to build Octopus Invaders from scratch. Qwen 3.5 9B running through Hermes Agent with full tool access on a single 3060 12GB. browser, file system, code execution, terminal. it reads the prompt, plans the architecture, writes every file, and serves the game. one prompt, zero steering, 11 minutes. i've been going back and forth with this model for days now. text, code, reasoning. it handles small chats, batching, bash scripting. at 9B it's fast enough to feel interactive and genuinely useful for local workflows where you'd normally reach for an API. but this is the real test. can it architect and build a full game autonomously without a single correction? it's coding right now. results coming next.

i gave a 9B model an agentic OS and told it to build Octopus Invaders from scratch. Qwen 3.5 9B running through Hermes Agent with full tool access on a single 3060 12GB. browser, file system, code execution, terminal. it reads the prompt, plans the architecture, writes every file, and serves the game. one prompt, zero steering, 11 minutes. i've been going back and forth with this model for days now. text, code, reasoning. it handles small chats, batching, bash scripting. at 9B it's fast enough to feel interactive and genuinely useful for local workflows where you'd normally reach for an API. but this is the real test. can it architect and build a full game autonomously without a single correction? it's coding right now. results coming next.

Sudo su

65,570 views • 6 months ago

The latest MLX has a CUDA back-end! To get started: pip install "mlx[cuda]" With the same codebase you can develop locally, run your model on Apple silicon, or in the cloud on Nvidia GPUs. MLX is designed around Apple silicon - which has a unified memory architecture. It uses the same design with CUDA. So there's no need to move arrays around from CPU memory to GPU memory. Note, this is early days - some operations are missing and performance is still being optimized. But it's already quite fast for Transformer training, text generation, and more! Here's a demo using mlx-lm to generate text with Llama 3 8B (bf16) on an A100:

The latest MLX has a CUDA back-end! To get started: pip install "mlx[cuda]" With the same codebase you can develop locally, run your model on Apple silicon, or in the cloud on Nvidia GPUs. MLX is designed around Apple silicon - which has a unified memory architecture. It uses the same design with CUDA. So there's no need to move arrays around from CPU memory to GPU memory. Note, this is early days - some operations are missing and performance is still being optimized. But it's already quite fast for Transformer training, text generation, and more! Here's a demo using mlx-lm to generate text with Llama 3 8B (bf16) on an A100:

Awni Hannun

42,810 views • 1 year ago

Introducing SmolVLM 256M (& 500M): The world’s smallest multimodal model. Designed for efficiency and perfect for on-device applications! 🔥 It’s so small, it can even run 100% locally in your browser on WebGPU! 🤏 Powered by Transformers.js! ⚡️ Try it out yourself! 👇

Introducing SmolVLM 256M (& 500M): The world’s smallest multimodal model. Designed for efficiency and perfect for on-device applications! 🔥 It’s so small, it can even run 100% locally in your browser on WebGPU! 🤏 Powered by Transformers.js! ⚡️ Try it out yourself! 👇

Xenova

25,302 views • 1 year ago

I tested Gemma 4 12b on my Macbook Pro M5 Max with 128GB of unified memory. It runs VERY slow for a 12b parameter model. 44 tokens/second. The lava lamp that it produced was also terrible. I would stick to Qwen 3.6 27b or Qwen 3.6 35b. Google is still way behind, even with their local models.

I tested Gemma 4 12b on my Macbook Pro M5 Max with 128GB of unified memory. It runs VERY slow for a 12b parameter model. 44 tokens/second. The lava lamp that it produced was also terrible. I would stick to Qwen 3.6 27b or Qwen 3.6 35b. Google is still way behind, even with their local models.

BridgeMind

70,626 views • 3 months ago

Microsoft has released its new open source model Phi-4 Incredibly powerful for its size and you can run it locally on your laptop! Completely free and available on Mac, Windows, Linux 🧵

Microsoft has released its new open source model Phi-4 Incredibly powerful for its size and you can run it locally on your laptop! Completely free and available on Mac, Windows, Linux 🧵

Paul Couvert

264,024 views • 1 year ago

So Alibaba Qwen has released the best image editing model... 100% open source! You can edit any photo using natural language. It can be used both locally and online. More below

So Alibaba Qwen has released the best image editing model... 100% open source! You can edit any photo using natural language. It can be used both locally and online. More below

Paul Couvert

246,616 views • 1 year ago

One more thing 🚀 Qwen’s agentic capability is no longer limited to the digital world — we’re bringing it into physical world. With our in-house robotic agentic system and navigation model, Qwen can now control a robot to execute tasks in real-world. #qwen #embodied #robotics

One more thing 🚀 Qwen’s agentic capability is no longer limited to the digital world — we’re bringing it into physical world. With our in-house robotic agentic system and navigation model, Qwen can now control a robot to execute tasks in real-world. #qwen #embodied #robotics

barry (xiong-hui) chen

25,112 views • 4 months ago

You can now run Qwen 3.6 27B on <10GB of memory and retain 90% of the intelligence. Today, PrismML dropped Bonsai 27B - which takes their unique compression approach they have already successfully applied to Qwen3.6 8B and Flux2-Klein and applied it to a properly smart model. Here is it using an entire computer for deep research. Long horizon agentic calls already are strong with Qwen3.6 27B and they still are here - just on 10x less memory.

You can now run Qwen 3.6 27B on <10GB of memory and retain 90% of the intelligence. Today, PrismML dropped Bonsai 27B - which takes their unique compression approach they have already successfully applied to Qwen3.6 8B and Flux2-Klein and applied it to a properly smart model. Here is it using an entire computer for deep research. Long horizon agentic calls already are strong with Qwen3.6 27B and they still are here - just on 10x less memory.

Tim Carambat

63,449 views • 2 months ago

WHY IS NO ONE TALKING ABOUT THIS?? Gemma 3n model was one of the best surprises for me. The fact that you can run it on edge devices even with just 2GB of RAM is impressive. A few weeks back, I was on holiday and used the Gemini Live feature a lot. But I kept running into issues whenever I was in a place where the network wasn’t reliable. Gemma 3n : > Multimodal: Supports text and image inputs; video/audio in development. > Context Window: Up to 128K tokens (32K for 1B model). > Multilingual: Trained on 140+ languages. > Privacy: Offline, on-device processing for data security. > Model Sizes: E2B (5B parameters, 2GB RAM) and E4B (10–12B parameters, 3GB RAM). Video credit: Google YT

WHY IS NO ONE TALKING ABOUT THIS?? Gemma 3n model was one of the best surprises for me. The fact that you can run it on edge devices even with just 2GB of RAM is impressive. A few weeks back, I was on holiday and used the Gemini Live feature a lot. But I kept running into issues whenever I was in a place where the network wasn’t reliable. Gemma 3n : > Multimodal: Supports text and image inputs; video/audio in development. > Context Window: Up to 128K tokens (32K for 1B model). > Multilingual: Trained on 140+ languages. > Privacy: Offline, on-device processing for data security. > Model Sizes: E2B (5B parameters, 2GB RAM) and E4B (10–12B parameters, 3GB RAM). Video credit: Google YT

AshutoshShrivastava

210,641 views • 1 year 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.

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 • 2 months ago