Benchmark Almanac

Vals Science Index

GPT-6 Astra 65.1% leads since Oct 2026 Not in model cards

Agents are scored on the equal average of BioMysteryBench, MysteryMechanism and Terminal-Bench-Science.

Version 1.0 released 2026-10-16. 23 models on 2026-10-25, led by GPT-6 Astra at 65.09%.

Made by Vals AI Released Oct 2026 Composite indices Tasks kept private Last model added Oct 2026 Scores refreshed 2 Nov 2026
Level
Expert Specialists with years of training, often over hours of work
Scored
Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
Answers
Agent task Many steps in a terminal, browser, desktop, app, game or robot.
Leaderboard
23
Models scored on its own leaderboard

GPT-6 Astra leads with 65.1%

Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

23 models. Scroll sideways for the rest.

  1. 65.1 GPT-6 Astra
  2. 59.6 GPT-6.1 Sol
  3. 59.1 Claude Sonnet 5.5
  4. 58.6 Claude Opus 5.5
  5. 55.4 Gemini 4 Argon
  6. 48.5 Muse Wonder Full
  7. 47.9 Claude Opus 5
  8. 45.0 GPT-6 Sol
  9. 44.4 Kimi K3.1 Pro
  10. 41.5 GPT-5.6 Sol
  11. 39.1 Muse Wonder
  12. 38.5 Claude Haiku 5.5
  13. 38.1 Grok 4.6
  14. 35.8 Gemini 3.8 Flash
  15. 34.8 Grok 4.7
  16. 31.8 Ember-1
  17. 31.7 MiMo-V2.6 Flash
  18. 29.6 DeepSeek V4.1 Flash
  19. 29.6 Step 5 Preview
  20. 28.4 GPT-6 Luna
  21. 27.4 Gemini 3.1 Pro
  22. 25.3 GPT-5.6 Luna
  23. 22.0 Beam 1.0
  1. 1 GPT-6 Astra 65.1%
  2. 2 GPT-6.1 Sol 59.6%
  3. 3 Claude Sonnet 5.5 59.1%
  4. 4 Claude Opus 5.5 58.6%
  5. 5 Gemini 4 Argon 55.4%
  6. 6 Muse Wonder Full 48.5%
  7. 7 Claude Opus 5 47.9%
  8. 8 GPT-6 Sol 45%
  9. 9 Kimi K3.1 Pro 44.4%
  10. 10 GPT-5.6 Sol 41.5%
  11. 11 Muse Wonder 39.1%
  12. 12 Claude Haiku 5.5 38.5%
  13. 13 Grok 4.6 38.1%
  14. 14 Gemini 3.8 Flash 35.8%
  15. 15 Grok 4.7 34.8%
  16. 16 Ember-1 31.8%
  17. 17 MiMo-V2.6 Flash 31.7%
  18. 18 DeepSeek V4.1 Flash 29.6%
  19. 19 Step 5 Preview 29.6%
  20. 20 GPT-6 Luna 28.4%
  21. 21 Gemini 3.1 Pro 27.4%
  22. 22 GPT-5.6 Luna 25.3%
  23. 23 Beam 1.0 22%
#ModelScoreSettingDateSource
1 GPT-6 Astra 65.1% Sep 2026 Official leaderboard
2 GPT-6.1 Sol 59.6% Sep 2026 Official leaderboard
3 Claude Sonnet 5.5 59.1% Sep 2026 Official leaderboard
4 Claude Opus 5.5 58.6% Sep 2026 Official leaderboard
5 Gemini 4 Argon 55.4% Sep 2026 Official leaderboard
6 Muse Wonder Full 48.5% Oct 2026 Official leaderboard
7 Claude Opus 5 47.9% Jul 2026 Official leaderboard
8 GPT-6 Sol 45% Sep 2026 Official leaderboard
9 Kimi K3.1 Pro 44.4% Oct 2026 Official leaderboard
10 GPT-5.6 Sol 41.5% Jul 2026 Official leaderboard
11 Muse Wonder 39.1% Oct 2026 Official leaderboard
12 Claude Haiku 5.5 38.5% Oct 2026 Official leaderboard
13 Grok 4.6 38.1% Aug 2026 Official leaderboard
14 Gemini 3.8 Flash 35.8% Sep 2026 Official leaderboard
15 Grok 4.7 34.8% Sep 2026 Official leaderboard
16 Ember-1 31.8% Sep 2026 Official leaderboard
17 MiMo-V2.6 Flash 31.7% Sep 2026 Official leaderboard
18 DeepSeek V4.1 Flash 29.6% Sep 2026 Official leaderboard
19 Step 5 Preview 29.6% Sep 2026 Official leaderboard
20 GPT-6 Luna 28.4% Sep 2026 Official leaderboard
21 Gemini 3.1 Pro 27.4% Feb 2026 Official leaderboard
22 GPT-5.6 Luna 25.3% Jul 2026 Official leaderboard
23 Beam 1.0 22% Oct 2026 Official leaderboard

The best score, 65.1%, costs $8.03 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 Xiaomi Corp Cost frontier One model at each effort level
0% 20% 40% 60% 80% $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6 Astra GPT-6.1 Sol MiMo-V2.6 Flash GPT-6 Luna 0% 20% 40% 60% 80% $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6 Astra GPT-6.1 Sol
ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6 Astra 65.1% $8.03
GPT-6.1 Sol 59.6% $1.36
Claude Sonnet 5.5 59.1% $8.92
Claude Opus 5.5 58.6% $9.13
Gemini 4 Argon 55.4% $18
Muse Wonder Full 48.5% $10
Claude Opus 5 47.9% $13
GPT-6 Sol 45% $2.31
Kimi K3.1 Pro 44.4% $3.33
GPT-5.6 Sol 41.5% $3.40
Muse Wonder 39.1% $8.74
Claude Haiku 5.5 38.5% $1.86
Grok 4.6 38.1% $2.45
Gemini 3.8 Flash 35.8% $2.98
Grok 4.7 34.8% $7.12
Ember-1 31.8% $11
MiMo-V2.6 Flash 31.7% $0.12
DeepSeek V4.1 Flash 29.6% $0.29
Step 5 Preview 29.6% $1.78
GPT-6 Luna 28.4% $0.11
Gemini 3.1 Pro 27.4% $2.56
GPT-5.6 Luna 25.3% $0.26
Beam 1.0 22%

Number one over time

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

OpenAI
Own leaderboard 1 leader
GPT-6 Astra 65.1%

Scores over time

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

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) GPT-6 Astra Claude Opus 5 GPT-5.6 Sol Gemini 3.1 Pro 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) GPT-6 Astra Claude Opus 5

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.

More in Other

All 41 in Other
  • Personal Superintelligence Index (Meta) In use Meta's composite of seven personal and knowledge-work evals: Agents' Last Exam, AutomationBench-AA, FocusBench, JobBench, GDP.pdf, VBT and DRACO.
  • Artificial Analysis Intelligence Index v5 In use Version 5 of Artificial Analysis's index: nine evaluations in four equal categories, most with private test sets, run three times each in its own agent harness. Gemini 4 Carbon 59.0
  • Anthropic ECI (AECI) Only in system cards Index that combines a basket of capability benchmarks into one capability score to track trends across Anthropic models. Claude Opus 5.5 169.4
  • Qwen in-house English automatic evaluation Only in system cards Models answer Qwen in-house English datasets on knowledge, comprehension, coding, math and related skills. Skills and weights change between reports. Qwen2-72B 72.9%
  • Recognition of model-written user turns (Anthropic) Only in system cards Models write synthetic user turns for real Claude Code transcripts, then judge which version is real. Score is accuracy above chance. Claude Mythos Preview 0.35
  • LiveBench Dropped Models answer monthly refreshed questions across math, coding, reasoning, language, instruction following, and data analysis. Scored automatically against objective ground truth. Claude Fable 5.1 85.7%