Autoform-Hard leaderboard
As of 19 Nov 2026, the top score on Autoform-Hard is 87.5% by GPT-6.2 Astra (OpenAI), ahead of Claude Fable 5.2 at 85.7%, across 8 model configurations with a published result (Vals AI, result dated 18 Nov 2026). Formalising graduate- and research-level mathematics in Lean 4: the 56 hardest tasks of AutoformBench, proofs checked by a clean-room rebuild. Run by Vals AI. Autoform-Hard is shown for reference only and does not count toward the BenchLeader Index.
Key facts (as of 19 Nov 2026)
- GPT-6.2 Astra holds the top Autoform-Hard score at 87.5% (Vals AI, 18 Nov 2026).
- Claude Fable 5.2 is second on Autoform-Hard at 85.7% (Vals AI, 18 Nov 2026).
- Autoform-Hard: Formalising graduate- and research-level mathematics in Lean 4: the 56 hardest tasks of AutoformBench, proofs checked by a clean-room rebuild. Run by Vals AI (Vals AI).
- 8 model configurations have a published Autoform-Hard result on BenchLeader (BenchLeader, 19 Nov 2026).
- Autoform-Hard does not feed the BenchLeader Index; it is shown for reference (BenchLeader methodology).
Formalising graduate- and research-level mathematics in Lean 4: the 56 hardest tasks of AutoformBench, proofs checked by a clean-room rebuild. Run by Vals AI.
- Published by
- Vals AI
- Category
- Maths
- Index weight
- Reference only
- Models
- 8
- Data as of
- 19 Nov 2026
Vals AI (vals.ai).
- 1GPT-6.2 Astra (max)87.5%
- 2Claude Fable 5.2 (high)85.7%
- 3GPT-6.2 Sol (max)85.7%
- 4Gemini 4 Carbon (high)71.4%
- 5Gemini 4 Hydrogen (xhigh)64.3%
- 6Gemini 4 Argon (high)48.2%
- 7Beam 1.00.0%
- 8MiniMax M3.1 Flash (max)0.0%
| # | |||||
|---|---|---|---|---|---|
| 1 | New | 87.5% | 68.9 | $15.24 | 2026-11-18 |
| 2 | 85.7% | 67.3 | $23.21 | 2026-11-18 | |
| 3 | New | 85.7% | 64.6 | $8.09 | 2026-11-18 |
| 4 | 71.4% | 68.0 | $12.38 | 2026-11-18 | |
| 5 | New | 64.3% | 63.6 | $23.25 | 2026-11-18 |
| 6 | 48.2% | 64.6 | $25.74 | 2026-11-18 | |
| 7 | 0.0% | 45.1 | – | 2026-11-18 | |
| 8 | 0.0% | – | $4.29 | 2026-11-18 |
AI reasoning and math benchmark leaderboard · All LLM benchmarks
Related benchmarks
Top head-to-heads on Autoform-Hard
Cite as: BenchLeader, “Autoform-Hard leaderboard”, data as of 19 Nov 2026, https://www.benchleader.com/benchmarks/vals_autoform_hard.
Autoform-Hard: questions
- What does Autoform-Hard measure?
- Formalising graduate- and research-level mathematics in Lean 4: the 56 hardest tasks of AutoformBench, proofs checked by a clean-room rebuild. Run by Vals AI. Scores are reported in percent of tasks solved; higher is better.
- What is the top score on Autoform-Hard?
- As of 19 Nov 2026, the highest score on Autoform-Hard is 87.5% by GPT-6.2 Astra (result dated 18 Nov 2026), ahead of Claude Fable 5.2 at 85.7%. Scores come from Vals AI and are re-read daily.
- Which AI model leads Autoform-Hard?
- GPT-6.2 Astra leads Autoform-Hard with 87.5% as of 19 Nov 2026 (result dated 18 Nov 2026), ahead of Claude Fable 5.2 at 85.7%.
- How many models have Autoform-Hard results?
- 8 model configurations have a Autoform-Hard result on BenchLeader, all taken from Vals AI.
- Who runs Autoform-Hard and how often is it updated?
- Autoform-Hard is published by Vals AI. BenchLeader re-reads the published results once a day and records the date each result was published.
- Does Autoform-Hard count toward the BenchLeader Index?
- No. Autoform-Hard is shown for reference but left out of the composite index.