G01na2 Gh1as1

Research Scientist at Google DeepMind. Co-creator of Gemini DeepThink. AI IMO Gold Medal 🥇

[Thang Luong @lmthang]

Continuing our IMO-gold journey, I’m delighted to share our #EMNLP2025 paper “Towards Robust Mathematical Reasoning”, which tells some of the key stories behind the success of our advanced Gemini #DeepThink at this year IMO. Finding the right north-star metrics was highly critical for our IMO effort and we did it with #IMOBench, a suite of advanced reasoning benchmarks for foundation models. More importantly, we encourage the community to go beyond short answers and showed that automatic grading of long-form answers is promising! Read on to see our project page, paper, and datasets in the thread 🙂

[Quoc Le @quocleix]

(1/3) Thrilled to announce a new Gemini breakthrough! Building on our success at IMO this year, an advanced version of Gemini Deep Think achieved gold-medal level performance at the ICPC 2025 World Finals - one of the world’s leading competitive programming competitions.
https://t.co/kDO4BkfqCP

[Thang Luong @lmthang]

Our IMO journey continues: the yolo run model that we trained a week before #imo2025, despite all possible likelihood of failures, magically achieves SOTA across a wide range of reasoning tasks from maths, to coding, and challenging knowledge. I'm very excited that we have now delivered the IMO 🥇 system to the hands of mathematicians and a simplified version (results below) to all Google AI Ultra subscribers.

[G01na2 Gh1as1 @g01na2]

DeepThink is officially out! 🚀

It’s been an incredible journey from our announcement at I/O to achieving Gold 🥇 at the IMO. We continuously pushed the boundaries, and we're thrilled to release a faster version that still secures an IMO Bronze 🥉.

[Google DeepMind @GoogleDeepMind]

For researchers, scientists, and academics tackling hard problems: Gemini 2.5 Deep Think is here. 🤯

It doesn't just answer, it brainstorms using parallel thinking and reinforcement learning techniques. We put it into the hands of mathematicians who explored what it can do ↓

[Thang Luong @lmthang]

Right before #imo2025, together with colleagues from Mountain View, NYC, Singapore, etc, we all gathered at @GoogleDeepMind headquarter in London for our final push for IMO. I believe that week was when all magic happened!

We put all individual recipes (that we figured out before) together and did a yolo run (with the compute that I had to beg various groups to loan) to train our most advanced Gemini model. We finished training 2 days before IMO :D That model achieved SOTA results, not just for math, but coding along with other reasoning tasks, unbelievable!

That week was also when we figured all details to scale our Deep Think mode, alongside with other enhanced inference strategies. We finalized our runbooks that predetermined the configs to be used during the IMO days (so there will be no human intervention at all). And the rest is history.

Thanks @CadeMetz and @nytimes for featuring that historic moment
https://t.co/b2TnTByzJ3

The photo below was taken in London during that magical week. Tagging various team members in random order who worked extremely for that moment @YiTayML @theophaneweber @DawsenHwang @JiemingMao @jon_lee0 @zichengxu42 @vinayramasesh @mirrokni @blackhc @gjb_ai @g01na2 @LeiYu63 @NateKushman @jj_at_brown @quocleix @freeetext together with many others that I don't yet have X handles.

[G01na2 Gh1as1 @g01na2]

It was an absolute blast helping bring Gemini DeepThink to life, and I'm beyond proud of what we've all accomplished! A year ago, we couldn't believe it was possible to get a gold medal with a general-purpose, text-in-text-out model!

[koray kavukcuoglu @koraykv]

Advanced version of Gemini Deep Think (announced at #GoogleIO) using parallel inference time computation achieved gold-medal performance at IMO, solving 5/6 problems with rigorous proofs as verified by official IMO judges! Congrats to all involved!

https://t.co/K41O8PUTDd

[Thang Luong @lmthang]

Last but not least, we experimented with a Gemini-based language reasoning system that showed great promise at this year’s IMO problems. This system doesn’t require the problems to be translated into a formal language and can directly generate and verify solutions in human-readable form.

This system is of great complement to AlphaGeometry and AlphaProof, as it demonstrated great potentials for the two hard combinatorics problems at #imo2024. The photo showed our team’s happiness when seeing some of the solutions! (guess what, the photo has 5 IMO medalists with a total of 5 🥇3 🥈)

[G01na2 Gh1as1 @g01na2]

Excited to share that our AI at Google DeepMind has achieved the equivalent of a silver medal at the International Mathematical Olympiad (#IMO2024). Proud to be part of this incredible accomplishment!

[Google DeepMind @GoogleDeepMind]

We’re presenting the first AI to solve International Mathematical Olympiad problems at a silver medalist level.🥈

It combines AlphaProof, a new breakthrough model for formal reasoning, and AlphaGeometry 2, an improved version of our previous system. 🧵 https://t.co/SYaLPSbIyj