Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

Bonsai 27B just changed the local LLM game forever. 1-bit quantization shrinks it from 54GB to just 3.8GB (-93%), while retaining 90% of its intelligence. That's insane. With custom WebGPU kernels written by Fable 5 and GPT 5.6 Sol, the model now runs locally in your browser!

xenovacom's profile picture

Xenova

20,596 subscribers

210,659 Aufrufe • vor 1 Monat •via X (Twitter)

Wissenschaft & Technologie
LIVE

Anya Rossi• Live Now

Private livecam show

0 Kommentare

Keine Kommentare verfügbar

Kommentare vom Original-Post werden hier angezeigt

Ähnliche Videos

It's official: Qwen3.8-27B just scored 52 on the Artificial Analysis Intelligence Index. We now have an open-weight model that matches GPT-5.6 Luna (max) AND can run locally... even in your browser with custom WebGPU kernels! What a time to be alive! 🤯
0:58

It's official: Qwen3.8-27B just scored 52 on the Artificial Analysis Intelligence Index. We now have an open-weight model that matches GPT-5.6 Luna (max) AND can run locally... even in your browser with custom WebGPU kernels! What a time to be alive! 🤯

Xenova

173,953 Aufrufe • vor 15 Tagen

A 27B parameter model used to need a server room. Now it runs on: • iPhone • Android • Mac PrismML's Bonsai makes it possible. • 1-bit weights • 27B params in just 3.9GB • ~90% of full precision quality (PrismML evals) • Better than the 2-bit version at less than half the size The entire Bonsai family is now live in RunAnywhere. • 1.7B to 27B models • 1-bit + 2-bit ternary • Thinking mode On iOS and Mac, Bonsai runs through ggml's llama.cpp and Apple's MLX. On Android, Bonsai runs through ggml's llama.cpp and directly on the Qualcomm Hexagon NPU through QHexRT, our proprietary inference runtime. We built custom silicon kernels to make 1-bit inference on an NPU possible for the first time. Two years ago this needed a data center. Now it thinks in airplane mode. Now available in the RunAnywhere app on the App Store and Google Play.
0:27

A 27B parameter model used to need a server room. Now it runs on: • iPhone • Android • Mac PrismML's Bonsai makes it possible. • 1-bit weights • 27B params in just 3.9GB • ~90% of full precision quality (PrismML evals) • Better than the 2-bit version at less than half the size The entire Bonsai family is now live in RunAnywhere. • 1.7B to 27B models • 1-bit + 2-bit ternary • Thinking mode On iOS and Mac, Bonsai runs through ggml's llama.cpp and Apple's MLX. On Android, Bonsai runs through ggml's llama.cpp and directly on the Qualcomm Hexagon NPU through QHexRT, our proprietary inference runtime. We built custom silicon kernels to make 1-bit inference on an NPU possible for the first time. Two years ago this needed a data center. Now it thinks in airplane mode. Now available in the RunAnywhere app on the App Store and Google Play.

RunAnywhere (YC W26)

16,892 Aufrufe • vor 1 Monat

this is f**king insane Fable 5 goes away in 24 hours from Claude Code and someone just figured out how to get fable 5 reasoning in GPT 5.6 Sol with one prompt you have less than 24 hours to set this up as this takes away 20% of your usage limit here is how: 1) Paste the prompt and ask it to "make an operating manual" 2) Go the Codex App > Settings > Personalisation > Custom Instructions 3) Use GPT 5.5 / GPT 5.6 and now you have Fable 5 reasoning at 1/3 th the price save and bookmark it no matter what
0:32

this is f**king insane Fable 5 goes away in 24 hours from Claude Code and someone just figured out how to get fable 5 reasoning in GPT 5.6 Sol with one prompt you have less than 24 hours to set this up as this takes away 20% of your usage limit here is how: 1) Paste the prompt and ask it to "make an operating manual" 2) Go the Codex App > Settings > Personalisation > Custom Instructions 3) Use GPT 5.5 / GPT 5.6 and now you have Fable 5 reasoning at 1/3 th the price save and bookmark it no matter what

Avid

57,544 Aufrufe • vor 1 Monat

Kimi K3 beating Fable and Sol on price and quality. Command Code's internal flappy bench results are in. Really impressed with the results. For Kimi K3 we set the gameplay to hard while Fable 5 and GPT-5.6 Sol were set to easier. Cost breakdown: • Kimi K3: $0.024 •Fable 5: $0.42 •GPT-5.6 Sol: $0.15 Ranking based on DX & features: •Kimi K3: 8/10 •Fable 5: 7.5/10 • GPT-5.6 Sol: 7/10 GPT 5.6 Sol max can definitely go on longer and wins the game overall, but that's something different. Grok 4.5 beats GPT 5.6 Sol on that, lol.Mm-hmm.
0:29

Kimi K3 beating Fable and Sol on price and quality. Command Code's internal flappy bench results are in. Really impressed with the results. For Kimi K3 we set the gameplay to hard while Fable 5 and GPT-5.6 Sol were set to easier. Cost breakdown: • Kimi K3: $0.024 •Fable 5: $0.42 •GPT-5.6 Sol: $0.15 Ranking based on DX & features: •Kimi K3: 8/10 •Fable 5: 7.5/10 • GPT-5.6 Sol: 7/10 GPT 5.6 Sol max can definitely go on longer and wins the game overall, but that's something different. Grok 4.5 beats GPT 5.6 Sol on that, lol.Mm-hmm.

Ahmad Awais

255,237 Aufrufe • vor 1 Monat

Previewing Ultrafast mode for OpenAI's GPT 5.6 Sol, powered by Cerebras. GPT-5.6 Sol Ultrafast generates responses at up to 750 tokens per second. That's the full, GPT-5.6-Sol model -- up to 14× faster than the same model on Standard processing. It speedran Humanity’s Last Exam in 11h 11m, nearly 7× faster than Claude Fable 5 with comparable accuracy.
0:46

Previewing Ultrafast mode for OpenAI's GPT 5.6 Sol, powered by Cerebras. GPT-5.6 Sol Ultrafast generates responses at up to 750 tokens per second. That's the full, GPT-5.6-Sol model -- up to 14× faster than the same model on Standard processing. It speedran Humanity’s Last Exam in 11h 11m, nearly 7× faster than Claude Fable 5 with comparable accuracy.

Cerebras

1,284,054 Aufrufe • vor 19 Tagen

Kimi K3 just 3 shotted a whole walkable voxel town with drivable cars — Voxelville, around $3.24 in API usage. The same token cost would be $10.80 with Fable 5 and $6 with GPT 5.6 Sol. Fable 5 built its own Maplewood, GPT 5.6 Sol built Willowwick, but on price/quality Kimi K3 just cooked both of them. The era of free indie game development is closer than you think!
0:28

Kimi K3 just 3 shotted a whole walkable voxel town with drivable cars — Voxelville, around $3.24 in API usage. The same token cost would be $10.80 with Fable 5 and $6 with GPT 5.6 Sol. Fable 5 built its own Maplewood, GPT 5.6 Sol built Willowwick, but on price/quality Kimi K3 just cooked both of them. The era of free indie game development is closer than you think!

Kirill

65,087 Aufrufe • vor 1 Monat

WTF?! This changes image generation forever! 🤯 PrismML just released Binary and Ternary Bonsai Image 4B! That's right, 1-bit diffusion models are here. Only ~3GB in size (FLUX.2 Klein 4B is 16GB). The most shocking part? It can run 100% locally in your browser. Try it now! 👇
0:37

WTF?! This changes image generation forever! 🤯 PrismML just released Binary and Ternary Bonsai Image 4B! That's right, 1-bit diffusion models are here. Only ~3GB in size (FLUX.2 Klein 4B is 16GB). The most shocking part? It can run 100% locally in your browser. Try it now! 👇

Xenova

183,996 Aufrufe • vor 3 Monaten

Before Fable 5 was shut down, it pushed Gemma 4 to 255 tok/s on WebGPU. Some didn't believe it was real. Today we're releasing the demo and kernels it wrote for you to see yourself. Run it locally in your browser. Agentic kernel optimization is the future of on-device inference
0:57

Before Fable 5 was shut down, it pushed Gemma 4 to 255 tok/s on WebGPU. Some didn't believe it was real. Today we're releasing the demo and kernels it wrote for you to see yourself. Run it locally in your browser. Agentic kernel optimization is the future of on-device inference

Xenova

484,504 Aufrufe • vor 2 Monaten

Another WTF moment. Using Codex as a fully custom game engine With the newly discovered and *insane* img2threejs skill 1) GPT imageGen a 3D item you want 2) ask GPT 5.6 Sol to use img2threejs 3) my harness brings it into the game automatically Just watch how good this is
0:31

Another WTF moment. Using Codex as a fully custom game engine With the newly discovered and *insane* img2threejs skill 1) GPT imageGen a 3D item you want 2) ask GPT 5.6 Sol to use img2threejs 3) my harness brings it into the game automatically Just watch how good this is

Nicolas Zullo

267,097 Aufrufe • vor 1 Monat

oh my.. this shouldn't be possible a 1B model that runs inside your browser, beats every model its size, and comes with its own desktop pet. MiniCPM-5 1B just changed the game for on-device AI. here's everything you need to know 🧵
0:32

oh my.. this shouldn't be possible a 1B model that runs inside your browser, beats every model its size, and comes with its own desktop pet. MiniCPM-5 1B just changed the game for on-device AI. here's everything you need to know 🧵

Muhammad Ayan

62,050 Aufrufe • vor 3 Monaten

Fable 5 just completely redesigned the website and the results are absolutely INSANE. It created a lightning bolt animation that strikes the Get Started CTA. It built an interactive demo of BridgeSpace from scratch. A perfect 1:1 of the real app. It created 3 marketing videos, one for each vibe coding tool in the BridgeMind ecosystem. Fable 5 is by far the best model at UI design. I don't know if GPT 5.6 can beat this.
0:52

Fable 5 just completely redesigned the website and the results are absolutely INSANE. It created a lightning bolt animation that strikes the Get Started CTA. It built an interactive demo of BridgeSpace from scratch. A perfect 1:1 of the real app. It created 3 marketing videos, one for each vibe coding tool in the BridgeMind ecosystem. Fable 5 is by far the best model at UI design. I don't know if GPT 5.6 can beat this.

BridgeMind

53,728 Aufrufe • vor 2 Monaten

GPT-5.6 vs GPT-5.5 on my custom spaceship prompt. I gave both models the exact same custom prompt. This is also the same prompt I previously gave to Fable 5. For context, GPT-5.6 Pro worked for 87 minutes, while GPT-5.5 Extra High worked for 34 minutes and 42 seconds. As I’ve said before, based on great authority GPT-5.6 will be an incremental/soldi improvement over GPT-5.5, not a “Fable killer.” My rough expectation has been that it would trade blows with Fable 5 on some benchmarks, maybe win around half depending on the category, but not clearly surpass it overall. And again fable five will have bigger model smell, but this was expected. After testing this coding output, that view feels pretty accurate. GPT-5.6 is clearly better than GPT-5.5 in several visual areas. The lighting, shading, chairs, object details, and exterior of the spaceship looked noticeably stronger. The scene was also easier to test. I do want to give GPT-5.5 credit though. It built out the rooms much much better and the planets looked better than GPT-5.6’s. It was also interesting that both GPT-5.5 and GPT-5.6 produced better-looking planets than Fable 5 in this specific test. The downside with GPT-5.5 was stability. The game was much glitchier and harder to test compared to GPT-5.6. But when it comes to the core of the demo, which is the spaceship itself, Fable 5 still beat both models pretty comfortably. GPT-5.6 is impressive, but from this test, it looks exactly like what I expected which was a meaningful incremental improvement over GPT-5.5, at least for indie game demos, but not something that replaces Fable 5. In collaboration with Chetaslua
0:42

GPT-5.6 vs GPT-5.5 on my custom spaceship prompt. I gave both models the exact same custom prompt. This is also the same prompt I previously gave to Fable 5. For context, GPT-5.6 Pro worked for 87 minutes, while GPT-5.5 Extra High worked for 34 minutes and 42 seconds. As I’ve said before, based on great authority GPT-5.6 will be an incremental/soldi improvement over GPT-5.5, not a “Fable killer.” My rough expectation has been that it would trade blows with Fable 5 on some benchmarks, maybe win around half depending on the category, but not clearly surpass it overall. And again fable five will have bigger model smell, but this was expected. After testing this coding output, that view feels pretty accurate. GPT-5.6 is clearly better than GPT-5.5 in several visual areas. The lighting, shading, chairs, object details, and exterior of the spaceship looked noticeably stronger. The scene was also easier to test. I do want to give GPT-5.5 credit though. It built out the rooms much much better and the planets looked better than GPT-5.6’s. It was also interesting that both GPT-5.5 and GPT-5.6 produced better-looking planets than Fable 5 in this specific test. The downside with GPT-5.5 was stability. The game was much glitchier and harder to test compared to GPT-5.6. But when it comes to the core of the demo, which is the spaceship itself, Fable 5 still beat both models pretty comfortably. GPT-5.6 is impressive, but from this test, it looks exactly like what I expected which was a meaningful incremental improvement over GPT-5.5, at least for indie game demos, but not something that replaces Fable 5. In collaboration with Chetaslua

Chris

250,919 Aufrufe • vor 2 Monaten

Introducing autoresearch with GPT 5.6 We had GPT 5.6 Sol reproduce the key findings from “Towards Mechanistically Understanding Why Memorized Knowledge Fails to Generalize in LLM Finetuning” Compared to GPT-5.5 and even Fable 5, GPT-5.6 stayed more focused on a few, critical experiments and spent less time on peripheral details. It also asked fewer “clarification” questions and independently resolved ambiguities instead of pushing them back to us OpenAI pushing the boundaries of the automated research loop with models that don’t have handcuffs
0:23

Introducing autoresearch with GPT 5.6 We had GPT 5.6 Sol reproduce the key findings from “Towards Mechanistically Understanding Why Memorized Knowledge Fails to Generalize in LLM Finetuning” Compared to GPT-5.5 and even Fable 5, GPT-5.6 stayed more focused on a few, critical experiments and spent less time on peripheral details. It also asked fewer “clarification” questions and independently resolved ambiguities instead of pushing them back to us OpenAI pushing the boundaries of the automated research loop with models that don’t have handcuffs

alphaXiv

88,164 Aufrufe • vor 1 Monat

The era of 1-bit LLMs is here — now with WebGPU acceleration! 🤯 It's incredible to think that a quantized 1.7B model (just 290MB in size) can run at ~100 tokens per second entirely in your browser. Try the demo yourself 👇
0:33

The era of 1-bit LLMs is here — now with WebGPU acceleration! 🤯 It's incredible to think that a quantized 1.7B model (just 290MB in size) can run at ~100 tokens per second entirely in your browser. Try the demo yourself 👇

Xenova

105,511 Aufrufe • vor 4 Monaten

this is f**king gold someone just figured out how to run Claude Fable 5 and GPT 5.6 Sol in the same loop gpt-5.6 came out yesterday. fable 5 leaves in 48 hours. this window has never existed before and never will again. here is how to use it: 1. open claude code. install openai's codex plugin 2. type /model. choose fable 5. 3. drop one ROUTING.md in your repo root (full file in comments). 4.paste the prompt (in comments), write what feature you want. boom. save and bookmark this no matter what. it will be the most productive thing you do this weekend
0:20

this is f**king gold someone just figured out how to run Claude Fable 5 and GPT 5.6 Sol in the same loop gpt-5.6 came out yesterday. fable 5 leaves in 48 hours. this window has never existed before and never will again. here is how to use it: 1. open claude code. install openai's codex plugin 2. type /model. choose fable 5. 3. drop one ROUTING.md in your repo root (full file in comments). 4.paste the prompt (in comments), write what feature you want. boom. save and bookmark this no matter what. it will be the most productive thing you do this weekend

Avid

140,354 Aufrufe • vor 1 Monat

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!
0:24

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 Aufrufe • vor 1 Jahr

Local models just got a free intelligence upgrade. Someone took the new Qwen3.8-27B (the one that already runs in 22GB) and only changed the chat template. No fine-tune. No new training. Just a smarter system prompt that kills all the filler and forces the model to stay sharp. Result: the same weights are now fixing real post-cutoff bugs faster and cleaner than Claude Opus 5 High in their tests. They call it the Sharp template. It’s free. Drop it on any Qwen3.5 / 3.6 / 3.8 model and it just works with llama.cpp. This is the kind of quiet upgrade that actually moves the needle for people running models locally.
0:39

Local models just got a free intelligence upgrade. Someone took the new Qwen3.8-27B (the one that already runs in 22GB) and only changed the chat template. No fine-tune. No new training. Just a smarter system prompt that kills all the filler and forces the model to stay sharp. Result: the same weights are now fixing real post-cutoff bugs faster and cleaner than Claude Opus 5 High in their tests. They call it the Sharp template. It’s free. Drop it on any Qwen3.5 / 3.6 / 3.8 model and it just works with llama.cpp. This is the kind of quiet upgrade that actually moves the needle for people running models locally.

Harman

73,661 Aufrufe • vor 14 Tagen

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.
0:24

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

62,879 Aufrufe • vor 1 Monat

NEW: LiquidAI just released LFM2.5-350M, a tiny model that brings agentic AI and tool-calling capabilities to resource-constrained environments. 🤯 It can even run locally in your browser via WebGPU, serving as a powerful companion while you browse the web. Try the demo! 👇
0:29

NEW: LiquidAI just released LFM2.5-350M, a tiny model that brings agentic AI and tool-calling capabilities to resource-constrained environments. 🤯 It can even run locally in your browser via WebGPU, serving as a powerful companion while you browse the web. Try the demo! 👇

Xenova

26,267 Aufrufe • vor 5 Monaten

Opus 5 High might be the most ridiculous model I’ve ever tested. It produced this result in around ONE HOUR. Fable 5 needed almost 3 hours. GPT-5.6 Sol worked for over an hour and still created almost nothing compared to this. Opus 5 didn’t just win. It completely OUTCLASSED them.
0:38

Opus 5 High might be the most ridiculous model I’ve ever tested. It produced this result in around ONE HOUR. Fable 5 needed almost 3 hours. GPT-5.6 Sol worked for over an hour and still created almost nothing compared to this. Opus 5 didn’t just win. It completely OUTCLASSED them.

OmedTheVibeCoder

39,695 Aufrufe • vor 1 Monat