Benchmark Almanac

AutoformBench

GPT-6 Astra 82.4% leads since October 2026
Not in model cards

Agents fill in missing Lean proofs for mathematical statements and submit their solutions to a clean grading workspace.

The release includes 341 audited tasks and model results with task costs.

Made by
Meta FAIR
Released
October 2026
Kind
Formal proofs
Level
Expert
Scored
Checked automatically
Answers
Written proof
Tasks
All public
Measures
Capability
Subject
Mathematics
Language
English
In and out
Text, Code → Code, Actions
Interaction
Agentic
Environment
Terminal, Code sandbox
Refresh
One fixed set
Run by
Anyone
In views
Agents
Strongest model tested
GPT-6 Astra #4 on the capability index
Last model added
September 2026
Scores refreshed
November 20th, 2026
9
Models scored on its own leaderboard

GPT-6 Astra leads with 82.4% among models released by September 2026

Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-6 Astra source-present 82.4%
  2. 2 Claude Opus 5 source-present 61.9%
  3. 3 GPT-5.6 Sol source-present 58.9%
  4. 4 Claude Fable 5 source-present 53.1%
  5. 5 DeepSeek V4.1 Flash source-present 52.8%
  6. 6 Gemini 3.7 Flash source-present 52.5%
  7. 7 Kimi K3 source-present 48.4%
  8. 8 Claude Sonnet 5 source-present 44.6%
  9. 9 Claude Opus 4.8 source-present 44%

Source: Official AutoformBench leaderboard

#ModelScoreSettingDateLinkSource
1 GPT-6 Astra 82.4% source-present Sep 2026 raw.githubusercontent.com Official leaderboard
2 Claude Opus 5 61.9% source-present Jul 2026 raw.githubusercontent.com Official leaderboard
3 GPT-5.6 Sol 58.9% source-present Jul 2026 raw.githubusercontent.com Official leaderboard
4 Claude Fable 5 53.1% source-present Jun 2026 raw.githubusercontent.com Official leaderboard
5 DeepSeek V4.1 Flash 52.8% source-present Sep 2026 raw.githubusercontent.com Official leaderboard
6 Gemini 3.7 Flash 52.5% source-present Aug 2026 raw.githubusercontent.com Official leaderboard
7 Kimi K3 48.4% source-present Jul 2026 raw.githubusercontent.com Official leaderboard
8 Claude Sonnet 5 44.6% source-present Jun 2026 raw.githubusercontent.com Official leaderboard
9 Claude Opus 4.8 44% source-present May 2026 raw.githubusercontent.com Official leaderboard

Rankings by part

We read 24 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as algebraic_combinatorics, algebraic_geometry_i and algebraic_topology_i. GPT-6 Astra leads 16 of them, and 4 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

algebraic_combinatorics

9 models we track have a score here.

  1. 1 Claude Opus 4.8 ↑8 100%
  2. 2 Claude Fable 5 ↑2 100%
  3. 3 GPT-5.6 Sol 100%
  4. 4 Kimi K3 ↑3 100%
  5. 5 Claude Opus 5 ↓3 100%
  6. 6 GPT-6 Astra ↓5 100%
  7. 7 Claude Sonnet 5 ↑1 50%
  8. 8 Gemini 3.7 Flash ↓2 50%
  9. 9 DeepSeek V4.1 Flash ↓4 50%

The best score, 82.4%, costs $43 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI DeepSeek Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $10 $20 $50 $100 Cost per task, log scale GPT-6 Astra, source-present DeepSeek V4.1 Flash, source-present 0% 20% 40% 60% 80% 100% $10 $100 Cost per task, log scale GPT-6 Astra, source-present DeepSeek V4.1 Flash, source-present

Source: Official AutoformBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6 Astra source-present 82.4% $43 36,958
Claude Opus 5 source-present 61.9% $44 639,377
GPT-5.6 Sol source-present 58.9% $44 60,790
Claude Fable 5 source-present 53.1% $97 420,444
DeepSeek V4.1 Flash source-present 52.8% $12 8,223,154
Gemini 3.7 Flash source-present 52.5% $17 380,530
Kimi K3 source-present 48.4% $42 343,017
Claude Sonnet 5 source-present 44.6% $42 1,094,963
Claude Opus 4.8 source-present 44% $27 396,627

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Own leaderboard Not tested since Oct 2026
GPT-6 Astra 82.4%

Source: Official AutoformBench leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-6 Astra Claude Opus 5 Claude Fable 5 Claude Opus 4.8 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-6 Astra Claude Opus 5

Source: Official AutoformBench leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Math

All 140 in Math
  • AIME 2026 In use Models solve the 30 integer-answer problems from the 2026 AIME I and II math competitions. Scored by exact answer match. GPT-5.5 100%
  • IMO-AnswerBench In use 400 olympiad problems with short verifiable answers, robustified against memorization. Scored by answer match. GLM-5.2 91%
  • HMMT 2026 In use Models solve the 33 final-answer problems from the February 2026 Harvard-MIT Mathematics Tournament. Scored by exact answer match. GPT-5.5 98.5%
  • PolyMath In use Models solve 9,000 parallel math problems in 18 languages across four difficulty levels, from school math to frontier math. Final answers are checked. Qwen3.7-Max 86.5%
  • MathArena (Apex, IMO, USAMO) In use Evaluates models on newly released math competitions after release. Final-answer or human/LLM-judged proof scores. Claude Opus 4.8 81.3%
  • MathArena Apex Shortlist In use Models solve 47 hard final-answer problems that MathArena selected from 2025 competitions, where the best models scored about 50% at selection. Scored by accuracy over repeated attempts. GPT-5.5 98.4%

Data audit

All 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (9 entries).
  • Pass Models scored 9 models we track.
  • Pass Newest score The newest is from September 2026, for DeepSeek V4.1 Flash.
  • Note Frontier models Only 1 of the top 10 has a score. The strongest tested is GPT-6 Astra (#4).
  • Pass Headroom The best is 82.4%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on github.com: 9 rows for models we track at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check