FormalTCS - Autoformalization: leaderboard
FormalTCS (2026) benchmarks LLMs on frontier theoretical computer science research over 143 instances from 2025-2026 conference papers, each annotated in natural language and in Lean 4. BEq+ equivalence rate for translating a natural-language theorem into a Lean 4 statement, verified by bidirectional proof search, in percent; higher is better.
Metric: BEq+ equivalence rate (%; natural-language theorem translated to a Lean 4 statement, counted when bidirectional proof search shows it equivalent to the reference statement; 8 sampled candidates per instance at temperature 0.6; 143 instances from theoretical computer science papers accepted at STOC, FOCS, SODA, COLT and similar venues in 2025-2026, each stage given the human-annotated input; each model runs in its own vendor harness). Source: arxiv.org. Saturation forecast: Around 2029. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 11.2 | #20 |
| 2 | GPT-5.6 Sol | 10.5 | #29 |
| 3 | Claude Sonnet 5 | 9.1 | #108 |
| 4 | GPT-5.6 Terra | 5.6 | #60 |
| 5 | GPT-5.6 Luna | 2.8 | #117 |
| 6 | Claude Haiku 4.5 | 2.1 | #568 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=formaltcs-autoformalization · How It Works · Data refreshed daily, snapshot 2026-11-12.