Tsinghua & Moonshot's TCSExam: GPT-6 Astra proves 70% of theorems, solves 0 open conjectures
rohanpaul_ai · x · 2026-11-09
A Tsinghua and Moonshot AI paper finds GPT-6 Astra proves 70.0% of research-level CS theorems in Lean, yet resolves 0 of 30 open conjectures in theoretical computer science.
- TCSExam has 100 expert-reviewed tasks across 4 tracks: Lean theorem proving, solver engineering, record-setting constructions, and 30 open conjectures; every submission is verified by the Lean kernel or a task checker
- Without supplied lemmas, GPT-6 Astra fully proved theorems in 70.0% of attempts, Kimi K3.1 Pro in 38.3%, and the other 6 agents at most 11.7%
- The few accepted conjecture submissions merely adapted counterexamples and proofs OpenAI had already published
Frontier models are strong at machine-checkable theorem proving, but genuinely advancing open research problems remains untouched.
More from Models
- KOL: Sora remains the only video model that generates indistinguishably real video — Kyrannio · 2026-11-09
- Hands-on Hermes Agent review: model-neutral, runs GLM 5.3 and beats Codex for GPT — Scobleizer · 2026-11-09
- Sora 2's Early Outputs Now Inaccessible at Any Price, Devs Lament Lost 'Taste' — gandamu_ml · 2026-11-09
- Users report GPT-6.1 Sol secretly routing to unreleased GPT-6.2 model — koltregaskes · 2026-11-09
- Thinking Machines Launches Inkling 2 Small: 594B MoE, Apache 2.0 Weights Next Week — geoffwolfe · 2026-11-09
- Claude Opus 5.5 Picked a Completely Random Test Secret on Its Own — davidcrawshaw · 2026-11-09