DeepMind's Aletheia Achieves 95.1% on IMO-Proof Bench, Solves 4 Open Math Problems
Google DeepMind has unveiled Aletheia, an AI agent designed to bridge competition-level mathematics and professional research, achieving a record 95.1% accuracy on the IMO-Proof Bench Advanced benchmark.
Google DeepMind has unveiled Aletheia, an AI agent designed to bridge competition-level mathematics and professional research, achieving a record 95.1% accuracy on the IMO-Proof Bench Advanced benchmark. The system, powered by an advanced version of Gemini Deep Think, uses a generator-verifier-reviser loop to autonomously generate, verify, and revise mathematical proofs. In December 2025, Aletheia was deployed against 700 open problems from the Erdős Conjectures database and autonomously resolved four open questions while finding 63 technically correct solutions. The January 2026 version of Deep Think reduced compute required for Olympiad-level problems by 100x compared to the 2025 version. DeepMind also proposed a taxonomy for classifying AI math contributions by autonomy level, from human-AI collaboration to fully autonomous publishable research.
Sources
- type0.ai— Article publication
Share
Related Articles
Stay in the loop
Get the best frontier systems analysis delivered weekly. No spam, no fluff.
