Very excited to share that an advanced version of Gemini Deep Think is the first to have achieved gold-medal level in the International Mathematical Olympiad! 🏆, solving five out of six problems perfectly, as verified by the IMO organizers! It’s been a wild run to lead this

"Optimizers Qualitatively Alter Solutions And We Should Leverage This" (arxiv.org/abs/2507.12224), very nice to see this direction of understanding what different optimizers bring in terms of solution properties.

🚀 We’re excited to introduce Qwen3-235B-A22B-Thinking-2507 — our most advanced reasoning model yet! Over the past 3 months, we’ve significantly scaled and enhanced the thinking capability of Qwen3, achieving: ✅ Improved performance in logical reasoning, math, science & coding

Proud to introduce Group Sequence Policy Optimization (GSPO), our stable, efficient, and performant RL algorithm that powers the large-scale RL training of the latest Qwen3 models (Instruct, Coder, Thinking) 🚀 📄 huggingface.co/papers/2507.18…

. Pablo Galindo Salgado 😂

I will be preparing some popcorn for what will be the new name of "context engineering" that was renamed a few days ago 😅

Many friends still ask me about AI for IMO, formal vs informal math. Some quick thoughts: IMO results: GDM and OpenAI achieved gold using informal (natural language) methods. ByteDance and AlphaProof (last year) got gold/silver using formal methods (Lean + specialized geometry

I think next years will tell that Yann LeCun was right on many accounts. If we look how discovery in science is unfolding today with world models (wm) + evolutionary methods on top of it, it is very similar to MPC as well.

I feel I can build an entire benchmark dataset with ONNX errors that would be harder than the humanity's last exam dataset for us to evaluate AGI

interesting swiglu variant from the gpt-oss model: clamps inputs and adds a skip connection

Amazing! Truly open review, through which we all gained more insights, i love it! Result: in multi epoch setting, making AR learn multiple orderings ~closes the gap to diffusion, explaining much of the difference. How the truly open review happened (from my vague memory): Mihir

Something fun we discovered: you can use #Genie3 to step into and explore your favorite paintings. Here's a short visit to Edward Hopper's "Nighthawks".

Oh wow, this VLM benchmark is pure evil, and I love it! "Vision Language Models are Biased" by An Vo, taesiri, Anh Totti Nguyen, etal. Also really good idea to have one-click copy-paste of images and prompts, makes trying it super easy.

I think the main question is: do we really need two separate AI for world modeling and another for agency ? I think these two concepts will merge very soon.

Introducing DINOv3: a state-of-the-art computer vision model trained with self-supervised learning (SSL) that produces powerful, high-resolution image features. For the first time, a single frozen vision backbone outperforms specialized solutions on multiple long-standing dense

Mesmerizing

Infinite monkey theorem. "No #shakespeare yet, but here's another copy of "The art of the deal"".

Not only shape but also utility.