Yang Chen

Research Scientist @NVIDIA | PhD @GeorgiaTech| RL and LLM reasoning

[Yang Chen @ychenNLP]

We released Nemotron Cascade 2 30B A3B.

What makes this release especially meaningful to me is that it reflects a 1.5-year journey at NVIDIA around one core idea: improving AI math reasoning through self-improvement at test time.

Each project tackled a different part of that problem.

With AceMath (24 Q4), we built an external verifier model to identify the right solution during test-time scaling.

With AceReason (24 Q1-2), we scaled the reasoning capabilities of the model through RL so the model could spend more time reflecting while solving problems. Along the way, we found a general, simple and effective RL recipe that we’ve kept using since.

And now with Cascade 2 (25 Q1), we’ve pushed that effort further: the model can generate hypotheses, verify them, and refine them on its own. That self-improvement loop is what enabled IMO gold-level performance at 30B level.

From MATH500, to AIME, and now IMO Proof.

This team is THE BEST.

Technical report:
https://t.co/V0XZn2ypPg

[Wei Ping @_weiping]

🚀 Introducing Nemotron-Cascade 2 🚀

Just 3 months after Nemotron-Cascade 1, we’re releasing Nemotron-Cascade 2: an open 30B MoE with 3B active parameters, delivering best-in-class reasoning and strong agentic capabilities.

🥇 Gold Medal-level performance on IMO 2025, IOI 2025, and ICPC World Finals 2025:
• Capabilities once thought achievable only by frontier proprietary models (e.g. Gemini Deep Think) or frontier-scale open models (i.e. DeepSeek-V3.2-Speciale-671B-A37B).
• Remarkably high intelligence density with 20× fewer parameters.

🏆 Best-in-class across math, code reasoning, alignment, and instruction following:
• Outperforms the latest Qwen3.5-35B-A3B (2026-02-24) and even larger Qwen3.5-122B-A10B (2026-03-11).

🧠 Powered by Cascade RL + multi-domain on-policy distillation:
• Significantly expand Cascade RL across a much broader range of reasoning and agentic domains than Nemotron-Cascade 1, while distilling from the strongest intermediate teacher models throughout training to recover regressions and sustain gains.

🤗 Model + SFT + RL data:
👉 https://t.co/4QJqfTOt6I

📄 Technical report:
👉 https://t.co/dFC00m6RZU

[zhiyang xu @zhiyangx11]

A 29B latent diffusion model, trained from scratch, grounded in physics, time, and space. Try it at https://t.co/limc0kT9n1

[Institute of Foundation Models @IFM_AI]

Introducing PAN 2: an open foundation world model for physical intelligence.

A language model predicts the next word. PAN 2 predicts the next world: what happens after each action, grounded in time, physics and space.

Towards AI that simulates “what if?” before it acts.

More details: https://t.co/lUX7igPOiF

[Yang Chen @ychenNLP]

@BillJohn1235813 @CSatUSC @robinomial @_jessethomason_ congrats!

[Cong Wei @CongWei1230]

Let's remove VAEs and ViTs from video models!

🚀 Introducing 𝗣𝗶𝘅𝗲𝗹𝗨𝗠𝗠: an encoder-free unified multimodal model for image and video understanding and generation, directly in pixel space.

Code and model available today!

🌐 https://t.co/RXUObhsSIZ
📄 https://t.co/lA7VaJoVws

[Yang Chen @ychenNLP]

@Kay2289123 补充一点：速度和吞吐不是一回事。单路 tok/s 是用户体验，成本看的是吞吐，这两是trade off关系。所以赚钱有两种模式：提吞吐降成本赚利润率或者把速度卖溢价

[Yang Chen @ychenNLP]

@ekzhang1 great post!
this makes a really clean case vs alternating prefill/decode passes. but I don't think it quite gets at Horace's q yet, "how PD compares to mixed chunk prefill". curious how you'd think about that one

[Yang Chen @ychenNLP]

@hexiang 🤣

[Yang Chen @ychenNLP]

@YangyiChen6666 🐶🐶🐶

[Yang Chen @ychenNLP]

@lu__jasper lol! I still remember when I tried to prompt the model with the hint and found the reasoning process completely changed after it.. my instinct told me this does not make sense for reasoning mdoels.. glad i stopped right there.

[Yang Chen @ychenNLP]

@ShuaichenChang @nvidia thank u 🙏🏻

[Yang Chen @ychenNLP]

Thrilled that Nemotron-Cascade has been accepted as an Oral at #NeurIPS2026! 🎉
It's our fully open-source RL recipe for LLM post-training. Huge thanks to all my amazing teammates at @NVIDIA.

model+data: https://t.co/XM7LWMz3Vd

[Yang Chen @ychenNLP]

🥈 Silver Medal at IOI 2025 & Outperforms DeepSeek-R1-0528 on LiveCodeBench.

Instead of mixing different tasks together, we scale *Cascade RL* to develop general LLMs in curriculum (RLFH -> Instruct -> Math -> Code -> SWE).
So many learnings, check out our report!👇

[Yang Chen @ychenNLP]

@CMHungSteven @nvidia 🤗

[Yang Chen @ychenNLP]

@chaojiang06 @nvidia 😀

[Jensen Huang @JensenHuang]

https://t.co/pppEbPVtfE

[Elon Musk @elonmusk]

SpaceX has committed to using Nvidia GPUs exclusively because they are the best

[Eric Zelikman @ericzelikman]

worth pointing out the nemotron folks have been doing exceptional open-weight model work for years now, and in general @nvidia has been a major supporter of many key open source efforts that all frontier labs have benefited from (directly or indirectly)

[Jensen Huang @JensenHuang]

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.

AI will transform every industry, power every company, and be built by every country.

Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.

The world needs both frontier closed models and frontier open models.

https://t.co/AUKzoQ5Ikb

[Hugging Apps @HuggingApps]

NVIDIA Nemotron just dropped an audio-native model that hears the world, not just words 🔥

transcription, translation, sound recognition & audio Q&A, TTS and full speech-to-speech, hears, thinks, and talks back natively

open weights, 2B and 30B

▶️ https://t.co/jUv9HVSfr3

[Cong Wei @CongWei1230]

💡LLM agents should do more than rewrite prompts—they should actively search for and organize useful multimodal context to support the generator.

Introducing SearchGen, which co-trains an LLM agent and a generator to expand the knowledge boundary of image generation.

Try our Live Demo: https://t.co/dlxoISD575
Search a frozen 1M Web corpus; no live API needed.

Read our project blog: https://t.co/3XxdjZ9R2j

(1/N)

[Yang Chen @ychenNLP]

@NielsRogge thx!

[Yang Chen @ychenNLP]

📢Excited to share a best-in-class end-to-end audio LLM that listen and speak with no loss in reasoning capability.

Building on Nemotron-Cascade-2, Audex tackles the core challenge that has bottlenecked audio model intelligence.

Check more details below!

[Wei Ping @_weiping]

🚀 Introducing Audex, a unified audio-text LLM for text, speech, sound, and music 🚀

🏆 Audex delivers best-in-class performance among open models across:
🎧 Audio understanding
🗣️ Speech recognition and translation
🔊 Text-to-speech
🎵 General audio generation
🔄 Speech-to-speech generation

🥇 Audex also achieves best-in-class results in math and code reasoning, alignment, and instruction following, even outperforming the text-only Qwen3.5-35B-A3B.

🧩 Minimalist architecture: a single 30B-A3B MoE model. Audio inputs are projected into the text embedding space, while text tokens and quantized audio output tokens are treated uniformly during generation.

🧠 Strong text intelligence, preserved
• Built on the Nemotron-Cascade-2-30B-A3B text backbone
• Trained with multi-stage SFT on blended audio-text data, Cascade RL, and multi-domain on-policy distillation
• The result: broad and SOTA audio capabilities with no regression in text intelligence.

🤗 Model:
👉 https://t.co/Plk2mJlARj

📄 Technical report:
👉 https://t.co/5Ic73VK7s2