Song Han retweeted
🚀 Open-H3-Regenerate-2K-IC: consistent video refinement up to 4K
We're releasing an improved version with much better consistency. The IC version keeps the low-res draft in context, so faces, lip sync, text, and layout stay faithful while high-res detail is generated.
🔌 Plug-and-play for MiniMax H3: just load the LoRA
⚙️ Flexible 3 / 4 / 8-step inference to balance speed and quality for your compute budget
🎬 2K / 4K output, plus 384P → 768P upscaling
📝 Project: nvlabs.github.io/Sana/Open-H…
💻 Code: github.com/NVlabs/Sana/tree/…
🤗 Weights: huggingface.co/Efficient-Lar…
Great work done by our team members @gao_sensen40956 & @HaoZhe65347
🚀 Open-H3-Regenerate-2K: 3-Step Video Regeneration, Native Output up to 4K
MiniMax open-sourced H3 Base for 768P video generation, but its official high-resolution regeneration stage, H3-Regenerate-2K, remains API-only. We provide an open alternative, bringing high-resolution H3 regeneration to your own GPUs.
⚙️ Just 3 diffusion steps, regenerating video directly at the target resolution
🎬 Native output up to 4K, supporting 768P, 1080P, 2K, and 4K for 5–15-second clips in one pass
🔌 Plug-and-play Refiner LoRA trained on H3: load the weights into H3 Base and start refining, with no retraining required
⚡ 5.0× speedup over the official API: for a 5-second 2K clip, the official API takes 112.7s; ours takes just 22.45s on a single B200, powered by SoL-Engine and Fast H3 VAE
📝 Blog: nvlabs.github.io/Sana/Open-H…
💻 Code: github.com/NVlabs/Sana/tree/…
Song Han retweeted
My lab has two papers at #CoRL2026 this week, and they make VLAs 2.4× and 4.7× faster.
FlashVLA does it for π0.5 on robot arms, and FlashDrive does it for NVIDIA's Alpamayo 1.5 in self-driving cars. That's fast enough for real-time control 🧵
Song Han retweeted
Proud of what our team has built in the past few months. Nunchux AI now leads 10 Artificial Analysis leaderboards for image and video generation speed, and many enterprise customers are running production image and video workloads on our stack. Our mission is to build the best and fastest infrastructure for image, video, and world models. If you build or run these models at scale, let's talk and work together.
Nunchux now leads 10 @ArtificialAnlys leaderboards for image and video generation speed, covering MiniMax-H3, LTX-2.3, Qwen-Image, and FLUX series. 🚀
These results come from building the Nunchux Model Optimizer and Nunchux Inference Engine together. We want to make image and video generation interactive and affordable, so you can try an idea and refine it as you go.
The waitlist is gone. Nunchux Modelverse is open to everyone, with 80+ models. More models and categories are on the way. Sign up today for free credits and generate faster for less.
Full results and comparisons on our blog: nunchux.ai/blog/nunchux-top-…
Special thanks to @ArtificialAnlys team for onboarding us to their benchmarks.
#Nunchux #AIInference #MultimodalAI #ImageGeneration #VideoGeneration
Song Han retweeted
Thanks to the Nebius AI Native Podcast team for having me!
From my research at @MIT to our work at @Eigen_AI_Labs, I’ve been focused on making AI faster and more efficient. What excites me most is how that work expands what developers can afford to build.
We talked about the engineering behind inference, why lower token costs can drive more usage, and when it makes sense for founders to build their own inference stack.
Enjoyed the conversation and hope you find it useful.
Watch at: youtu.be/w1MJV6GGH3I?si=TZk4…
A smaller open model can beat a frontier model on the right task.
On the AI Native Podcast, Ryan Wang discusses what he learned building @Eigen_AI_Labs, the economics of inference, and why cheaper tokens will expand the market. Listen here: nebius.com/podcast youtube.com/watch?v=w1MJV6GG…
Song Han retweeted
Our co-founder @zhijianliu on why the drafter becomes the bottleneck in speculative decoding, and how parallel drafting gets past it ⚡
DFlash 2 builds on this: about one more accepted token per verification pass (+21%), for just 1.3% more cycle latency.
A takeaway from my #Runtime talk recently: drafting one token at a time becomes its own bottleneck.
With #DFlash, we draft a whole block in parallel. That makes room for a deeper, more accurate drafter while keeping latency low. The target model still verifies every token.
At @inco_ai, we’re building on this work to make inference faster for AI applications and agents.
Song Han retweeted
Nunchux now leads 10 @ArtificialAnlys leaderboards for image and video generation speed, covering MiniMax-H3, LTX-2.3, Qwen-Image, and FLUX series. 🚀
These results come from building the Nunchux Model Optimizer and Nunchux Inference Engine together. We want to make image and video generation interactive and affordable, so you can try an idea and refine it as you go.
The waitlist is gone. Nunchux Modelverse is open to everyone, with 80+ models. More models and categories are on the way. Sign up today for free credits and generate faster for less.
Full results and comparisons on our blog: nunchux.ai/blog/nunchux-top-…
Special thanks to @ArtificialAnlys team for onboarding us to their benchmarks.
#Nunchux #AIInference #MultimodalAI #ImageGeneration #VideoGeneration
Song Han retweeted
🚀 Can making an LLM weaker actually make it smarter?
Excited to share our latest work: Weaken to Strengthen — Weakening-Induced On-Policy Self-Distillation (WISD)!
💡 Instead of relying on a stronger teacher, we deliberately weaken the model through quantization or pruning, then use the contrast to improve its reasoning.
No external teacher. No reference solutions. No reward signals. Just the model itself and unlabeled prompts!
🔥 Key results:
📈 +4.0 / +5.5 percentage points average accuracy gains with quantization / pruning.
🧮 HMMT 2025: 57.7% → 70.5% (Qwen3.5-4B).
💻 LiveCodeBench v6: 39.1% → 56.7% (Qwen3.5-4B).
Sometimes, getting stronger starts with getting weaker!
💻 Code: github.com/jet-ai-projects/W…
🌐 Website: jet-ai-projects.github.io/WI…
#LLM #Reasoning #SelfDistillation #EfficientAI
Song Han retweeted
🚀 Excited to share our latest work: Delta-Matching — Closing the Final Gap in Native 8-bit LLM Training!
Delta-Matching lets us train softmax attention with all seven core matmuls in FP8. In our experiments, it matches BF16/FP32 training loss and preserves overall downstream performance where evaluated.
Delta-Matching computes the correction from the same operands as the gradient. We derive the error introduced by stale delta and track selected channel gains growing before loss rises. In our GDN/GQA runs, larger models show faster gain growth; longer runs reveal degradation that short tests miss.
We tested
• Attention variants with and without QK normalization, RoPE and NoPE, and head dimensions 128 and 256.
• GDN/GQA, GDN/MLA, KDA/GQA, Mamba-2/GQA and pure-GQA model architectures.
The experiments span 569M to 5.29B parameters, 8K pretraining and lower-learning-rate context extension from 8K to 64K. We also tested AdamW and Muon-based training, different global batch sizes, and FP8 FFNs and attention projections.
🌐 Blog: jet-ai-projects.github.io/de…
📄 Paper: arxiv.org/abs/2609.37852
💻 Kernel Code: github.com/jet-ai-projects/d…
#EfficientAI #LLM #FP8 #DeepLearning #OpenSource
Song Han retweeted
🚀 Open-H3-Regenerate-2K: 3-Step Video Regeneration, Native Output up to 4K
📝 Blog: nvlabs.github.io/Sana/Open-H…
💻 Code: github.com/NVlabs/Sana/tree/…
🤗 HF Weights: huggingface.co/Efficient-Lar…
Song Han retweeted
After ICLR, we finally got some time to top first again on lean-eval leaderboard with updated Humanize workflows 🤗
> LeanEval is the last unsaturated benchmark in mathematical formalization.
> New Humanize flow learnt a lot of insights from OpenAI and Anthropic's blogs, and Axiom's tech report.
> More progress on "big name theorem" is coming :)
teapaw.org/IlinVasily29521/status…
LeanEval is the last unsaturated benchmark in mathematical formalization. Saturating it will require formalizing Fermat's Last Theorem. @axiommathai is currently in the lead.
Leaderboard: lean-lang.org/eval/legacy/
Song Han retweeted
Happy to share that our LongLive-Plug has been integrated into ComfyUI. Check it out at registry.comfy.org/publisher…
📄 Paper: arxiv.org/abs/2609.38154
💻 Code: github.com/NVlabs/LongLive/t…
🤗 Models: huggingface.co/collections/E…
Song Han retweeted
🚀 Introducing Robota-Embed — a unified Vision-Language-Action (VLA) embedding model that enables robots to retrieve and learn from past experiences.
Key Highlights:
🎯 Better Retrieval: 47.33% R@1, outperforming Gemini Embedding 2 by +22.09 points.
🤖 Better Robot Learning: Success rate improves from 47.0% → 56.3% on DROID-Sim and 86.4% → 92.8% on LIBERO.
⚡ Self-Evolving Agents: 24% → 68% task success with 43.7% fewer policy tokens on ManiSkill, compared to the direct GPT-6 Astra baseline.
The right experience. Better decisions. More efficient physical intelligence.
📄 Paper: arxiv.org/abs/2610.19495
🌐 Project: robota-ai.github.io/robota-e…
#Robotics #PhysicalAI #EmbodiedAI #VLA
Song Han retweeted
🚀 Introducing Lightning Weave — a new approach to composing capabilities from post-trained accuracy and efficiency specialists, creating reasoning models that are both more accurate and more efficient.
⚡ Results on Qwen3.5-4B:
HMMT: 59.2 → 64.0 (+4.8 points) with 10.7% fewer tokens
LiveCodeBench: 41.7 → 54.2 (+12.5 points) with 9.6% fewer tokens
🧠 Better reasoning. Fewer tokens.
📄 Website: jet-ai-projects.github.io/Li…
Song Han retweeted
Excited to release Long-WAM, our latest exploration in embodied AI 🤖 @AaronWeiHuang
Memory helps it track object motion, predict trajectories, and improve on long-horizon tasks as context grows. Our infra brings real-time control to edge devices.
Explore Long-WAM 👇
🌐 Project: nvlabs.github.io/LongLive/Lo…
📄 Paper: arxiv.org/abs/2610.10528
💻 Code: github.com/NVlabs/LongLive/t…
Five findings 🧵👇
1/ Longer context, better WAMs.
RoboCasa GR-1: 66.3% → 78.7% (2.4s → 19.2s). LIBERO-Long: 94.5% → 99.5%. Memory helps predict object motion.
2/ Pretraining matters.
LongLive2.0-Robot, pretrained on ~10,000 hours of robot/egocentric video, scales with context; the bidirectional baseline does not.
3/ Smoother async control.
RoboTwin 2.0: 94.2% async vs. 94.4% sync. Action jerk is ~3× lower than Fast-WAM and ~15× lower than LingBot-VA, without blending or prefix guidance.
4/ Real-time, edge-ready infra.
Optimized for RTX 5090, DGX Spark & AGX Thor. Just 107.4 ms per action chunk on RTX 5090, including IDM-style video prediction.
5/ Strong planners still need fast policies.
GPT-6 Astra plans; Long-WAM executes. RoboCasa365: 31.4% → 54.4%. Unseen composite tasks: 6.1% → 35.0%.
Song Han retweeted
🚀 Excited to introduce StreamRender-H3 — bringing real-time interactive worlds to life with diffusion models!
Built on Sol-Engine, we post-trained @MiniMax_AI H3 Ref2VA into a real-time streaming neural renderer.
🎮 Code drives the world. A game engine handles physics, object states, and player interactions, while H3 generates the visuals.
⚡ Real-time streaming: 2 denoising steps, 24 FPS, and ~273 ms per chunk on 8×GB200 GPUs.
🎨 One world, endless appearances: Change the reference image and prompt to render the same interactive world in entirely different styles.
🔥 Powered by streaming post-training, distillation, a fast causal H3 VAE, and full-stack inference acceleration.
Check out our demo and code! 👇
demo: nvlabs.github.io/Sana/Sol-En…
code: github.com/NVlabs/Sana/tree/…
Song Han retweeted
Excited to release Long-WAM, our new exploration in embodied AI 🤖！
With memory, Long-WAM precisely perceives how objects move, predicts their trajectories, and gains emergent scaling on long-horizon tasks; our infra runs it fast on edge devices.
nvlabs.github.io/LongLive/Lo…
Song Han retweeted
We’ve built a really strong team at @NunchuxAI , and we’re still hiring.
We’re looking for great people across product, R&D and GTM to join us for the next stage!
We're growing our SF team. Join us and build the frontier of multimodal inference at Nunchux AI.
Engineering
✦ Product Engineer
✦ ML Engineer, Visual Generative Models
✦ ML Engineer, Model Integrations
GTM
✦ Founding Marketer
✦ Founding GTM
nunchux.ai/careers
Song Han retweeted
Splash 1.3.0 is out, and it's about the wait before your local agent starts answering ⚡
If you switch projects and come back, SSD offloading brings the first token down from 19 seconds to 1.
New tasks start 1.5× faster, thanks to the Neural Engine. All on a 24 GB M6 Mac.
Song Han retweeted
Excited to share our latest NeurIPS 2026 work, Jet-Long — extending LLM context with no fine-tuning and ≤4% generation overhead, while delivering strong long-context accuracy! 🚀
🚀 Introducing Jet-Long — efficient long-context extension for LLMs with Dynamic Bifocal RoPE.
Extend Qwen3 from 32K → 128K, Qwen3-30B-A3B-Instruct from 256k → 1M context with:
• 🚫 No fine-tuning
• 📈 +4.79/+2.18/+2.03 pp on RULER (1.7B/4B/8B) over the strongest baseline
• 🧠 Best overall HELMET-RAG & lowest PG-19 perplexity
• ⚡ ≤4% generation overhead
Jet-Long dynamically adapts RoPE to the current sequence length while exactly preserving the base model within its pretrained context window.
Long context, essentially free at inference.
📄 Paper (NeurIPS 2026): arxiv.org/abs/2607.07740
💻 Github: github.com/jet-ai-projects/j…
Song Han retweeted
31× more video per GPU, at 1/15 the cost ⚡
That's @linewiseAI's video agent on Inco vs. their own pipeline.
The agent is thorough. Before it calls a moment idle, it checks what came before and after, reads the sensors, and zooms in. Lots of model calls, and we make them fast.
One week of our video data: 16,190 hours. A full-time worker would need 8 years to watch it.
Our agent-native infra analyzes it in a day, with inference on @inco_ai: 31× the throughput per GPU at 1/15 the cost per hour of video.