DataCanvas's Yanbiao Lab tops Terence Tao-backed math distillation challenge
At the end of September 2026, the SAIR Foundation co-founded by Terence Tao published results for the second stage of the Mathematics Distillation Challenge, where Yanbiao Lab (yanbiao.ai), part of AI infrastructure and computing-cloud provider DataCanvas, scored a full 200 points in all four tracks and topped the main leaderboard with 221 points; on the separate Research Leaderboard it led with 17 points against 3 for the runner-up. The competition requires systems to submit Lean 4 formal proofs when an implication holds and a counterexample when it does not, with every submission checked by a deterministic verifier, shifting the criterion from answer accuracy to verifiability. The team then published a paper that, among the 96 order-five candidate equations locked by ETP, constructs non-trivial infinite models for 28 equations that had no public classification, establishing 28 new Austin classifications in 14 dual classes, with one search producing 636 Lean 4 certificates that the verifier accepted.
If formal verification plus a reinforcement-learning loop can become reusable infrastructure, AI-for-science results can be auditable and traceable rather than just leaderboard scores.