Victor M's banner

Victor M

@victormustar • 28,574 subscribers

🤗 Head of Product @huggingface

Shorts

Fable has done AGI-level job on on the Boeing 747 benchmark... it's almost scary 👀

Fable has done AGI-level job on on the Boeing 747 benchmark... it's almost scary 👀

2,613,276 görüntüleme

some genius invented a kebab benchmark for llms

some genius invented a kebab benchmark for llms

317,893 görüntüleme

Before the week ends, let's acknowledge one of the most INSANE week ever for open AI, with 25+ notable open-weight drops across every modality: 🧠 LLMs → NVIDIA Nemotron 3 Ultra: 550B hybrid Mamba-MoE, only 55B active, 1M context, MMLU 89.1. NVFP4 variant claims ~5x throughput on Blackwell. First openly-weighted 550B hybrid Mamba-Transformer, closing the gap with frontier closed models. → Google Gemma 4 12B: fully open dense any-to-any (text/image/audio/video), 256k context, encoder-free, 140+ languages, AIME 2026 at 77.5. Shipped with a 23-checkpoint QAT wave (mobile ONNX + MLX). Most deployable model of the week. → StepFun Step-3.7-Flash: 198B sparse MoE VLM, ~11B active, SWE-Bench PRO 56.3. Apache 2.0. → Liquid AI LFM2.5-8B-A1B: edge MoE, just 1.5B active, 128k ctx, MATH500 88.8, MLX-ready. Best on-device option this week. → JetBrains Mellum2-12B-A2.5B-Thinking: their first open MoE, near-Qwen3-14B coding at 2.5B active. Apache 2.0. 🎨 Image gen (the surprise of the week) → Ideogram 4: their FIRST-EVER open weights. 9.3B flow-matching DiT trained from scratch. #2 overall behind GPT Image 2, top open-weight model on Design Arena + LMArena. Strongest open checkpoint for text-rich images, full stop. It has taste. Still can't believe this is open weights. 🔊 Audio & Speech (a breakout week for open TTS, 4 labs shipped) → Boson Higgs Audio v3 4B: 102 languages, 21 emotions, singing/whispering/shouting, sub-second TTFA. → RedNote dots.tts: the only fully continuous (no codec) open TTS pipeline, Apache 2.0. → Google Magenta RealTime 2: real-time music gen, <200ms latency, text+audio+MIDI. multimodalart ported it to PyTorch within hours with live ZeroGPU demos. → NVIDIA Nemotron-3.5 ASR: 600M streaming, 17x more concurrent streams vs Parakeet RNNT 1.1B. 👁️ Vision & VLMs → PaddleOCR-VL-1.6: SOTA document parsing at 1B params, Apache 2.0. → Baidu NAVA: 6.3B joint audio-video gen, best-in-class A/V sync, Apache 2.0. 🎬 Video, 3D & World Models → NVIDIA Cosmos3-Super: 64B omnimodal world model coupling action trajectories with video+audio gen, for Physical AI. → JD JoyAI-Echo: up to 5-min multi-shot text-to-video on LTX-2.3. → ByteDance Bernini-R + VAST TripoSplat (single-image-to-3D Gaussian splats, MIT).

Before the week ends, let's acknowledge one of the most INSANE week ever for open AI, with 25+ notable open-weight drops across every modality: 🧠 LLMs → NVIDIA Nemotron 3 Ultra: 550B hybrid Mamba-MoE, only 55B active, 1M context, MMLU 89.1. NVFP4 variant claims ~5x throughput on Blackwell. First openly-weighted 550B hybrid Mamba-Transformer, closing the gap with frontier closed models. → Google Gemma 4 12B: fully open dense any-to-any (text/image/audio/video), 256k context, encoder-free, 140+ languages, AIME 2026 at 77.5. Shipped with a 23-checkpoint QAT wave (mobile ONNX + MLX). Most deployable model of the week. → StepFun Step-3.7-Flash: 198B sparse MoE VLM, ~11B active, SWE-Bench PRO 56.3. Apache 2.0. → Liquid AI LFM2.5-8B-A1B: edge MoE, just 1.5B active, 128k ctx, MATH500 88.8, MLX-ready. Best on-device option this week. → JetBrains Mellum2-12B-A2.5B-Thinking: their first open MoE, near-Qwen3-14B coding at 2.5B active. Apache 2.0. 🎨 Image gen (the surprise of the week) → Ideogram 4: their FIRST-EVER open weights. 9.3B flow-matching DiT trained from scratch. #2 overall behind GPT Image 2, top open-weight model on Design Arena + LMArena. Strongest open checkpoint for text-rich images, full stop. It has taste. Still can't believe this is open weights. 🔊 Audio & Speech (a breakout week for open TTS, 4 labs shipped) → Boson Higgs Audio v3 4B: 102 languages, 21 emotions, singing/whispering/shouting, sub-second TTFA. → RedNote dots.tts: the only fully continuous (no codec) open TTS pipeline, Apache 2.0. → Google Magenta RealTime 2: real-time music gen, <200ms latency, text+audio+MIDI. multimodalart ported it to PyTorch within hours with live ZeroGPU demos. → NVIDIA Nemotron-3.5 ASR: 600M streaming, 17x more concurrent streams vs Parakeet RNNT 1.1B. 👁️ Vision & VLMs → PaddleOCR-VL-1.6: SOTA document parsing at 1B params, Apache 2.0. → Baidu NAVA: 6.3B joint audio-video gen, best-in-class A/V sync, Apache 2.0. 🎬 Video, 3D & World Models → NVIDIA Cosmos3-Super: 64B omnimodal world model coupling action trajectories with video+audio gen, for Physical AI. → JD JoyAI-Echo: up to 5-min multi-shot text-to-video on LTX-2.3. → ByteDance Bernini-R + VAST TripoSplat (single-image-to-3D Gaussian splats, MIT).

538,948 görüntüleme

Finally spent 45 minutes with Grok 3 👀 Impression so far: it's the best code model I've ever used. I kept throwing random ideas at it and it grew the project to 400 lines almost error-free.

Finally spent 45 minutes with Grok 3 👀 Impression so far: it's the best code model I've ever used. I kept throwing random ideas at it and it grew the project to 400 lines almost error-free.

6,635,093 görüntüleme

Funny how you can do the same thing with a local model without sending all your data to a remote cloud.

Funny how you can do the same thing with a local model without sending all your data to a remote cloud.

1,196,385 görüntüleme

300tok/s on mobile is insane... open source must win ✊

300tok/s on mobile is insane... open source must win ✊

93,457 görüntüleme

Reminder: every Hugging Face Space is an API your agents can call :) I asked mine to build a website about the flowers of France 🌸 and it used VAST AI's TripoSplat Space to turn photos it found into real 3D Gaussian splats, live on the page! All on my HF Pro daily ZeroGPU credits (40 min/day renewed daily for only $9/month)

Reminder: every Hugging Face Space is an API your agents can call :) I asked mine to build a website about the flowers of France 🌸 and it used VAST AI's TripoSplat Space to turn photos it found into real 3D Gaussian splats, live on the page! All on my HF Pro daily ZeroGPU credits (40 min/day renewed daily for only $9/month)

168,074 görüntüleme

NVIDIA's Kimodo is the release of the week 🔥 Prompt the timeline whatever your want like: "a person walks forward" → "a person starts jumping", hit Generate, and watch a 3D character do it in seconds (700hrs of pro mocap training. Works on human + robot skeletons. Super fast + free to use on HF)

NVIDIA's Kimodo is the release of the week 🔥 Prompt the timeline whatever your want like: "a person walks forward" → "a person starts jumping", hit Generate, and watch a 3D character do it in seconds (700hrs of pro mocap training. Works on human + robot skeletons. Super fast + free to use on HF)

419,155 görüntüleme

I just made my own Star Wars clip using AI (text-to-video)!

I just made my own Star Wars clip using AI (text-to-video)!

3,012,404 görüntüleme

Crazy what you can do on Hugging Face in 40 seconds

Crazy what you can do on Hugging Face in 40 seconds

437,595 görüntüleme

Why is a 100% open source chatbot faster than ChatGPT "Instant"? 👀

Why is a 100% open source chatbot faster than ChatGPT "Instant"? 👀

234,666 görüntüleme

Microsoft TRELLIS.2 is here 🔥 • Single image → textured 3D mesh • 4B params, flow-matching transformer • Up to 1536³ resolution • Open weights, MIT licensed ⬇️ Demo available on Hugging Face

Microsoft TRELLIS.2 is here 🔥 • Single image → textured 3D mesh • 4B params, flow-matching transformer • Up to 1536³ resolution • Open weights, MIT licensed ⬇️ Demo available on Hugging Face

194,494 görüntüleme

New Voxtral-4B-TTS is beautiful 💛 it sounds so good that it's whispering at closed models: "your time is over" - Demo available on Hugging Face ⬇️

New Voxtral-4B-TTS is beautiful 💛 it sounds so good that it's whispering at closed models: "your time is over" - Demo available on Hugging Face ⬇️

107,012 görüntüleme

Now that we have amazing open source TTS with fast inference, what are you building?

Now that we have amazing open source TTS with fast inference, what are you building?

431,129 görüntüleme

another open source win: opencode + GLM 4.6 is basically Claude Code (used it all day) but insanely cheap + better TUI. And you can use it with your Hugging Face token now 🔥

another open source win: opencode + GLM 4.6 is basically Claude Code (used it all day) but insanely cheap + better TUI. And you can use it with your Hugging Face token now 🔥

217,637 görüntüleme

New Llama.cpp UI is a blessing for the local AI world 🌎 - Blazing fast, beautiful, and private (ofc) - Use 150,000+ GGUF models in a super slick UI - Drop in PDFs, images, or text documents - Branch and edit conversations anytime - Parallel chats and image processing - Math and code rendering - Constrained generation with JSON schema supported Easy setup + open-source + community-built 🔥

New Llama.cpp UI is a blessing for the local AI world 🌎 - Blazing fast, beautiful, and private (ofc) - Use 150,000+ GGUF models in a super slick UI - Drop in PDFs, images, or text documents - Branch and edit conversations anytime - Parallel chats and image processing - Math and code rendering - Constrained generation with JSON schema supported Easy setup + open-source + community-built 🔥

161,184 görüntüleme

290MB = useful LLM (running directly in your browser via WebGPU) 🤯 Bonsai 1-bit is beyond my comprehension... try the demo on hugging face ⬇️

290MB = useful LLM (running directly in your browser via WebGPU) 🤯 Bonsai 1-bit is beyond my comprehension... try the demo on hugging face ⬇️

63,157 görüntüleme

Huge drop by Meta: ActionMesh turns any video into an animated 3D mesh. Demo available on Hugging Face ⬇️

Huge drop by Meta: ActionMesh turns any video into an animated 3D mesh. Demo available on Hugging Face ⬇️

94,543 görüntüleme

Mac owners don't miss this: MLX LM is now integrated directly within Hugging Face 🤯 ⬇️ Run 4,400+ LLMs locally on Apple Silicon at max speed, no cloud, no wait.

Mac owners don't miss this: MLX LM is now integrated directly within Hugging Face 🤯 ⬇️ Run 4,400+ LLMs locally on Apple Silicon at max speed, no cloud, no wait.

204,678 görüntüleme

I’m not saying an open-source model you can run on consumer hardware is better than ChatGPT, just that sometimes it’s worth trying both. Prompt: “A zebra with a volcanic eruption in the background.”

I’m not saying an open-source model you can run on consumer hardware is better than ChatGPT, just that sometimes it’s worth trying both. Prompt: “A zebra with a volcanic eruption in the background.”

105,329 görüntüleme

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

LTX-2 trainer is a huge deal. I tried it to add water to the podracers scene using their demo water-sim fine-tune (it's a fine-tune whose only goal is to add water :D) I think we're going to see film productions that don't use general models. They'll train their own fine-tunes, built for exactly what they need and consistent between shots for long content.

LTX-2 trainer is a huge deal. I tried it to add water to the podracers scene using their demo water-sim fine-tune (it's a fine-tune whose only goal is to add water :D) I think we're going to see film productions that don't use general models. They'll train their own fine-tunes, built for exactly what they need and consistent between shots for long content.

94,204 görüntüleme • 26 gün önce

llama.cpp with MTP support makes local models fast enough to use as daily drivers 🚀 Qwen3.6-27B dense generation (on A10G): From 25 tok/s → 45 tok/s (+78%). Two flags on llama-server: --spec-type draft-mtp --spec-draft-n-max 2

llama.cpp with MTP support makes local models fast enough to use as daily drivers 🚀 Qwen3.6-27B dense generation (on A10G): From 25 tok/s → 45 tok/s (+78%). Two flags on llama-server: --spec-type draft-mtp --spec-draft-n-max 2

171,753 görüntüleme • 2 ay önce

My last hours with Fable: I was building this movement parkour sim before it went down... Impressed by its autonomy: built its own self-verifying harness with its own rubric for how a movement should feel. When a new movement was added it could tell on its own what felt right or wrong until it felt 100% right without me in the loop (and it was very good at it) Fable was more than just another model iteration imo. For the short (but intense) time it was available, it felt like playing with clay: ideas became code with almost no friction and the line between both became blurry. More than ever: open source MUST win. I don't want a world where intelligence is centralized and you're stuck with a hand saw while others have a chainsaw.

My last hours with Fable: I was building this movement parkour sim before it went down... Impressed by its autonomy: built its own self-verifying harness with its own rubric for how a movement should feel. When a new movement was added it could tell on its own what felt right or wrong until it felt 100% right without me in the loop (and it was very good at it) Fable was more than just another model iteration imo. For the short (but intense) time it was available, it felt like playing with clay: ideas became code with almost no friction and the line between both became blurry. More than ever: open source MUST win. I don't want a world where intelligence is centralized and you're stuck with a hand saw while others have a chainsaw.

82,807 görüntüleme • 1 ay önce

GLM-5.1 > Claude Code (Opus 4.6)? I'm tripping or CC has become very bad but built a Three.js racing game to eval and it's extremely impressive. Thoughts: - One-shot car physics with real drift mechanics (this is hard) - My fav part: Awesome at self iterating (with no vision!) created 20+ Bun.WebView debugging tools to drive the car programmatically and read game state. Proved a winding bug with vector math without ever seeing the screen - 531-line racing AI in a single write: 4 personalities, curvature map, racing lines, tactical drifting. Built telemetry tools to compare player vs AI speed curves and data-tuned parameters - All assets from scratch: 3D models, procedural textures, sky shader, engine sounds, spatial AI audio! - Can do hard math: proved road normals pointed DOWN via vector cross products, computed track curvature normalized by arc length to tune AI cornering speed You are going to hear about this model a lot in the next months - open source let's go 🚀🚀

GLM-5.1 > Claude Code (Opus 4.6)? I'm tripping or CC has become very bad but built a Three.js racing game to eval and it's extremely impressive. Thoughts: - One-shot car physics with real drift mechanics (this is hard) - My fav part: Awesome at self iterating (with no vision!) created 20+ Bun.WebView debugging tools to drive the car programmatically and read game state. Proved a winding bug with vector math without ever seeing the screen - 531-line racing AI in a single write: 4 personalities, curvature map, racing lines, tactical drifting. Built telemetry tools to compare player vs AI speed curves and data-tuned parameters - All assets from scratch: 3D models, procedural textures, sky shader, engine sounds, spatial AI audio! - Can do hard math: proved road normals pointed DOWN via vector cross products, computed track curvature normalized by arc length to tune AI cornering speed You are going to hear about this model a lot in the next months - open source let's go 🚀🚀

193,488 görüntüleme • 3 ay önce

Watch Gemma-4-E4B casually identify sea animals by classifying images in a single agentic session using its vision capabilities. (impressive for a 4B model 🚀) I'm convinced: the agents of tomorrow are local, free, fast, and run on every computer!

Watch Gemma-4-E4B casually identify sea animals by classifying images in a single agentic session using its vision capabilities. (impressive for a 4B model 🚀) I'm convinced: the agents of tomorrow are local, free, fast, and run on every computer!

171,280 görüntüleme • 3 ay önce

Open source AI music is actually good now 🔥 Made a free demo for ACE-Step 1.5: describe any song, get it back in seconds ⤵️

Open source AI music is actually good now 🔥 Made a free demo for ACE-Step 1.5: describe any song, get it back in seconds ⤵️

99,588 görüntüleme • 3 ay önce

Why is a 100% open source chatbot faster so much faster then ChatGPT to turn an image into anime? 👀

Why is a 100% open source chatbot faster so much faster then ChatGPT to turn an image into anime? 👀

205,530 görüntüleme • 8 ay önce

This is wild 🤯 We finally have an open-source music generation model that rocks 🚀

This is wild 🤯 We finally have an open-source music generation model that rocks 🚀

248,949 görüntüleme • 1 yıl önce

it's open source time, with a real leap for world models 🎉 NVIDIA's SANA-WM: a camera-conditioned world model that fits on one GPU. 60s of 720p in 34s on a single 5090 - 2.6B params and Apache 2.0!

it's open source time, with a real leap for world models 🎉 NVIDIA's SANA-WM: a camera-conditioned world model that fits on one GPU. 60s of 720p in 34s on a single 5090 - 2.6B params and Apache 2.0!

34,386 görüntüleme • 2 ay önce

Why isn't everyone using FLUX.2-dev-Turbo? Top quality, great at editing, 1.5s/image speed, 120 images for $1. My go-to for AI image gen/editing now 🔥

Why isn't everyone using FLUX.2-dev-Turbo? Top quality, great at editing, 1.5s/image speed, 120 images for $1. My go-to for AI image gen/editing now 🔥

77,986 görüntüleme • 6 ay önce

New TTS bomb dropped on Hugging Face. It has multispeaker support and the output quality looks amazing 🤯

New TTS bomb dropped on Hugging Face. It has multispeaker support and the output quality looks amazing 🤯

112,585 görüntüleme • 1 yıl önce

🤯 Mind blown by the realism of podcasts generated by NotebookLM...

🤯 Mind blown by the realism of podcasts generated by NotebookLM...

165,958 görüntüleme • 1 yıl önce

DeepSite V2 just dropped (insane for Vibe Coding) 🤯 🔹Amazingly good at UI Design 🔹DeepSeek-R1 model now available 🔹Targeted Edits on any element 🔹Redesign any existing website 🔹100% open source (UI + models)

DeepSite V2 just dropped (insane for Vibe Coding) 🤯 🔹Amazingly good at UI Design 🔹DeepSeek-R1 model now available 🔹Targeted Edits on any element 🔹Redesign any existing website 🔹100% open source (UI + models)

102,584 görüntüleme • 1 yıl önce

Introducing: HuggingChat Omni 💫 Select the best model for every prompt automatically 🚀 - Automatic model selection for your queries - 115 models available across 15 providers Available now all Hugging Face users. 100% open source.

Introducing: HuggingChat Omni 💫 Select the best model for every prompt automatically 🚀 - Automatic model selection for your queries - 115 models available across 15 providers Available now all Hugging Face users. 100% open source.

72,106 görüntüleme • 9 ay önce

who made this 🇫🇷 banger lol (identify yourself)

who made this 🇫🇷 banger lol (identify yourself)

28,235 görüntüleme • 3 ay önce

Hugging Face Spaces, the world's largest AI app directory, is now MCP-compatible 🤯 Here turning an entire website into Ghibli in one shot with Claude Code to demonstrate what's now possible: instantly access any A I tool right in your LLM client!

Hugging Face Spaces, the world's largest AI app directory, is now MCP-compatible 🤯 Here turning an entire website into Ghibli in one shot with Claude Code to demonstrate what's now possible: instantly access any A I tool right in your LLM client!

82,163 görüntüleme • 1 yıl önce

Absolutely insane release from Sesame 🤯 ⬇️ try the HF online demo

Absolutely insane release from Sesame 🤯 ⬇️ try the HF online demo

71,554 görüntüleme • 1 yıl önce

Kimi-K2 ✖️ Z-Image-Turbo (god I love open source)

Kimi-K2 ✖️ Z-Image-Turbo (god I love open source)

34,278 görüntüleme • 7 ay önce

We're entering an era where small, specialized AI models beat huge general models at specific tasks. Example: TESSA-T1 nails React UI generation: clean state logic, smart components, clear design. Available on Hugging Face 🤗

We're entering an era where small, specialized AI models beat huge general models at specific tasks. Example: TESSA-T1 nails React UI generation: clean state logic, smart components, clear design. Available on Hugging Face 🤗

60,144 görüntüleme • 1 yıl önce

Announcing HF Spaces' biggest update 🤯 → Ask anything you want to do with AI. With ~400k Spaces, this is the best place to find the AI apps you need ✅

Announcing HF Spaces' biggest update 🤯 → Ask anything you want to do with AI. With ~400k Spaces, this is the best place to find the AI apps you need ✅

57,155 görüntüleme • 1 yıl önce