Benchmark Almanac

PolicyBench

GPT-6 Sol 95.2% leads since Sep 2026 Not in model cards

No-tools estimate of US taxes and benefits for 100 households across 18 programs. Headline metric is answers within $1 of PolicyEngine-US.

Prompts and reference values are public, so the authors call the board a preview. Forced tool calls lowered some Claude scores.

Made by PolicyEngine Released Jul 2026 Finance All tasks public 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
Short answer A number, word or short string matched to a key.
PolicyEngine/policybench 2 stars Paper Leaderboard
51
Entries on its own leaderboard, 50 models we track

GPT-6 Sol leads with 95.2%

Best score per entry 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. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

51 entries. Scroll sideways for the rest.

  1. 95.2 GPT-6 Sol
  2. 94.1 Claude Opus 5.5
  3. 93.1 GPT-6.1 Terra
  4. 93.1 GPT-5.6 Sol
  5. 91.7 GPT-6 Astra
  6. 91.7 Claude Sonnet 5.5
  7. 91.5 GPT-6 Luna
  8. 90.8 Claude Fable 5.1
  9. 90.5 Kimi K3
  10. 90.2 GPT-6.1 Sol
  11. 90.0 Muse Wonder
  12. 88.7 GPT-5.6 Luna
  13. 88.6 Gemini 4 Argon
  14. 88.6 ox-alpha
  15. 88.0 GPT-5.5
  16. 87.7 Inkling
  17. 87.6 Grok 4.7
  18. 87.6 GPT-5.6 Terra
  19. 87.2 Muse Spark 1.3
  20. 86.3 DeepSeek V4.1 Flash
  21. 85.8 Grok 4.6
  22. 85.6 Gemini 3.8 Flash
  23. 84.7 Grok 4.5
  24. 83.6 Claude Opus 5
  25. 83.3 Claude Fable 5
  26. 83.0 DeepSeek V4 Pro 0813
  27. 82.9 DeepSeek V4 Flash 0731
  28. 82.7 Gemini 3.7 Flash
  29. 82.6 Gemini 3.6 Flash
  30. 81.5 Gemini 3.5 Flash-Lite
  31. 81.2 Gemini 3.1 Pro
  32. 81.1 GLM-5.3
  33. 80.6 Gemini 3 Flash
  34. 80.5 Claude Opus 4.7
  35. 80.3 Grok 4.3
  36. 79.9 Claude Sonnet 4.6
  37. 79.9 Grok Build 0.1
  38. 79.8 Gemini 3.5 Flash
  39. 79.3 Gemini 3.1 Flash-Lite
  40. 79.3 DeepSeek-V4-Pro
  41. 79.2 Claude Haiku 5.5
  42. 76.9 Qwen3.7-Max
  43. 76.5 GLM-5.2
  44. 75.8 Claude Opus 4.8
  45. 75.4 MiniMax-M3
  46. 74.8 Qwen 3.8 Max
  47. 74.5 Claude Haiku 4.5
  48. 73.3 GPT-5.4 Mini
  49. 72.3 Claude Sonnet 5
  50. 66.9 Kimi K2.6
  51. 65.3 GPT-5.4 Nano
  1. 1 GPT-6 Sol 95.2%
  2. 2 Claude Opus 5.5 94.1%
  3. 3 GPT-6.1 Terra 93.1%
  4. 4 GPT-5.6 Sol 93.1%
  5. 5 GPT-6 Astra 91.7%
  6. 6 Claude Sonnet 5.5 91.7%
  7. 7 GPT-6 Luna 91.5%
  8. 8 Claude Fable 5.1 90.8%
  9. 9 Kimi K3 90.5%
  10. 10 GPT-6.1 Sol 90.2%
  11. 11 Muse Wonder 90%
  12. 12 GPT-5.6 Luna 88.7%
  13. 13 Gemini 4 Argon 88.6%
  14. 14 ox-alpha 88.6%
  15. 15 GPT-5.5 88%
  16. 16 Inkling 87.7%
  17. 17 Grok 4.7 87.6%
  18. 18 GPT-5.6 Terra 87.6%
  19. 19 Muse Spark 1.3 87.2%
  20. 20 DeepSeek V4.1 Flash 86.3%
  21. 21 Grok 4.6 85.8%
  22. 22 Gemini 3.8 Flash 85.6%
  23. 23 Grok 4.5 84.7%
  24. 24 Claude Opus 5 83.6%
  25. 25 Claude Fable 5 83.3%
  26. 26 DeepSeek V4 Pro 0813 83%
  27. 27 DeepSeek V4 Flash 0731 82.9%
  28. 28 Gemini 3.7 Flash 82.7%
  29. 29 Gemini 3.6 Flash 82.6%
  30. 30 Gemini 3.5 Flash-Lite 81.5%
  31. 31 Gemini 3.1 Pro 81.2%
  32. 32 GLM-5.3 81.1%
  33. 33 Gemini 3 Flash 80.6%
  34. 34 Claude Opus 4.7 80.5%
  35. 35 Grok 4.3 80.3%
  36. 36 Claude Sonnet 4.6 79.9%
  37. 37 Grok Build 0.1 79.9%
  38. 38 Gemini 3.5 Flash 79.8%
  39. 39 Gemini 3.1 Flash-Lite 79.3%
  40. 40 DeepSeek-V4-Pro 79.3%
  41. 41 Claude Haiku 5.5 79.2%
  42. 42 Qwen3.7-Max 76.9%
  43. 43 GLM-5.2 76.5%
  44. 44 Claude Opus 4.8 75.8%
  45. 45 MiniMax-M3 75.4%
  46. 46 Qwen 3.8 Max 74.8%
  47. 47 Claude Haiku 4.5 74.5%
  48. 48 GPT-5.4 Mini 73.3%
  49. 49 Claude Sonnet 5 72.3%
  50. 50 Kimi K2.6 66.9%
  51. 51 GPT-5.4 Nano 65.3%
#EntryScoreSettingDateSource
1 GPT-6 Sol 95.2% Sep 2026 Official leaderboard
2 Claude Opus 5.5 94.1% Sep 2026 Official leaderboard
3 GPT-6.1 Terra 93.1% Oct 2026 Official leaderboard
4 GPT-5.6 Sol 93.1% Jul 2026 Official leaderboard
5 GPT-6 Astra 91.7% Sep 2026 Official leaderboard
6 Claude Sonnet 5.5 91.7% Sep 2026 Official leaderboard
7 GPT-6 Luna 91.5% Sep 2026 Official leaderboard
8 Claude Fable 5.1 90.8% Sep 2026 Official leaderboard
9 Kimi K3 90.5% Jul 2026 Official leaderboard
10 GPT-6.1 Sol 90.2% Sep 2026 Official leaderboard
11 Muse Wonder 90% Oct 2026 Official leaderboard
12 GPT-5.6 Luna 88.7% Jul 2026 Official leaderboard
13 Gemini 4 Argon 88.6% Sep 2026 Official leaderboard
14 ox-alpha 88.6% Official leaderboard
15 GPT-5.5 88% Apr 2026 Official leaderboard
16 Inkling 87.7% Jul 2026 Official leaderboard
17 Grok 4.7 87.6% Sep 2026 Official leaderboard
18 GPT-5.6 Terra 87.6% Jul 2026 Official leaderboard
19 Muse Spark 1.3 87.2% Sep 2026 Official leaderboard
20 DeepSeek V4.1 Flash 86.3% Sep 2026 Official leaderboard
21 Grok 4.6 85.8% Aug 2026 Official leaderboard
22 Gemini 3.8 Flash 85.6% Sep 2026 Official leaderboard
23 Grok 4.5 84.7% Jul 2026 Official leaderboard
24 Claude Opus 5 83.6% Jul 2026 Official leaderboard
25 Claude Fable 5 83.3% Jun 2026 Official leaderboard
26 DeepSeek V4 Pro 0813 83% Aug 2026 Official leaderboard
27 DeepSeek V4 Flash 0731 82.9% Jul 2026 Official leaderboard
28 Gemini 3.7 Flash 82.7% Aug 2026 Official leaderboard
29 Gemini 3.6 Flash 82.6% Jul 2026 Official leaderboard
30 Gemini 3.5 Flash-Lite 81.5% Jul 2026 Official leaderboard
31 Gemini 3.1 Pro 81.2% Feb 2026 Official leaderboard
32 GLM-5.3 81.1% Aug 2026 Official leaderboard
33 Gemini 3 Flash 80.6% Dec 2025 Official leaderboard
34 Claude Opus 4.7 80.5% Apr 2026 Official leaderboard
35 Grok 4.3 80.3% Apr 2026 Official leaderboard
36 Claude Sonnet 4.6 79.9% Feb 2026 Official leaderboard
37 Grok Build 0.1 79.9% May 2026 Official leaderboard
38 Gemini 3.5 Flash 79.8% May 2026 Official leaderboard
39 Gemini 3.1 Flash-Lite 79.3% Mar 2026 Official leaderboard
40 DeepSeek-V4-Pro 79.3% Apr 2026 Official leaderboard
41 Claude Haiku 5.5 79.2% Oct 2026 Official leaderboard
42 Qwen3.7-Max 76.9% May 2026 Official leaderboard
43 GLM-5.2 76.5% Jun 2026 Official leaderboard
44 Claude Opus 4.8 75.8% May 2026 Official leaderboard
45 MiniMax-M3 75.4% Jun 2026 Official leaderboard
46 Qwen 3.8 Max 74.8% Aug 2026 Official leaderboard
47 Claude Haiku 4.5 74.5% Oct 2025 Official leaderboard
48 GPT-5.4 Mini 73.3% Mar 2026 Official leaderboard
49 Claude Sonnet 5 72.3% Jun 2026 Official leaderboard
50 Kimi K2.6 66.9% Apr 2026 Official leaderboard
51 GPT-5.4 Nano 65.3% Mar 2026 Official leaderboard

The best score, 95.2%, costs $0.03 a task

Runs on the cost frontier score higher than any run that's cheaper. GPT-6 Luna comes within 3.7 points of the best for $0.002 a task, 14 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Google DeepMind OpenAI Anthropic Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 Cost per task, log scale GPT-6 Sol GPT-6.1 Terra GPT-6 Luna Gemini 3.1 Flash-Lite 0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 Cost per task, log scale GPT-6 Sol Gemini 3.1 Flash-Lite
ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6 Sol 95.2% $0.03
Claude Opus 5.5 94.1% $0.07
GPT-6.1 Terra 93.1% $0.01
GPT-5.6 Sol 93.1% $0.09
GPT-6 Astra 91.7% $0.13
Claude Sonnet 5.5 91.7% $0.03
GPT-6 Luna 91.5% $0.002
Claude Fable 5.1 90.8% $0.26
Kimi K3 90.5% $0.47
GPT-6.1 Sol 90.2% $0.02
Muse Wonder 90% $0.08
GPT-5.6 Luna 88.7% $0.02
Gemini 4 Argon 88.6% $0.03
GPT-5.5 88% $0.26
Inkling 87.7% $0.09
Grok 4.7 87.6% $0.26
GPT-5.6 Terra 87.6% $0.04
Muse Spark 1.3 87.2% $0.02
DeepSeek V4.1 Flash 86.3% $0.02
Grok 4.6 85.8% $0.09
Gemini 3.8 Flash 85.6% $0.07
Grok 4.5 84.7% $0.07
Claude Opus 5 83.6% $0.07
Claude Fable 5 83.3% $0.54
DeepSeek V4 Pro 0813 83% $0.20
DeepSeek V4 Flash 0731 82.9% $0.003
Gemini 3.7 Flash 82.7% $0.01
Gemini 3.6 Flash 82.6% $0.05
Gemini 3.5 Flash-Lite 81.5% $0.003
Gemini 3.1 Pro 81.2% $0.08
GLM-5.3 81.1% $0.17
Gemini 3 Flash 80.6% $0.05
Claude Opus 4.7 80.5% $0.29
Grok 4.3 80.3% $0.02
Claude Sonnet 4.6 79.9% $0.19
Grok Build 0.1 79.9% $0.04
Gemini 3.5 Flash 79.8% $0.06
Gemini 3.1 Flash-Lite 79.3% $0.002
DeepSeek-V4-Pro 79.3% $0.01
Claude Haiku 5.5 79.2% $0.001
Qwen3.7-Max 76.9% $0.14
GLM-5.2 76.5% $0.08
Claude Opus 4.8 75.8% $0.26
MiniMax-M3 75.4% $0.007
Qwen 3.8 Max 74.8% $0.22
Claude Haiku 4.5 74.5% $0.06
GPT-5.4 Mini 73.3% $0.006
Claude Sonnet 5 72.3% $0.17
Kimi K2.6 66.9% $0.39
GPT-5.4 Nano 65.3% $0.002

3 models have held number one since its release in Jul 2026

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

OpenAI
Own leaderboard Passed 85% in Jul 2026, by GPT-5.5
GPT-6 Sol 95.2%

The best score passed 85% in Apr 2026

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 Sol GPT-5.5 Gemini 3.1 Pro Claude Haiku 4.5 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) GPT-6 Sol GPT-5.5

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 Law, finance and medicine

All 140 in Law, finance and medicine
  • HealthBench Professional In use 525 physician-authored tasks from real clinician chats (care consult, documentation, research) with physician-adjudicated rubrics. Claude Sonnet 5.5 69.2%
  • MedXpertQA-MM In use Models answer expert-level medical exam questions that combine images with clinical records and test results. Scoring is accuracy. Qwen 3.8 Max 80.4%
  • HealthBench In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
  • KBL In use Models answer Korean legal knowledge and reasoning tasks and Korean bar exam questions. It has 7 knowledge tasks, 4 reasoning tasks and 2,510 bar exam items. Solar Pro 4 77.5%
  • KorMedMCQA In use Models answer 7,469 multiple-choice questions from Korean licensing exams for doctors, nurses, pharmacists and dentists, held from 2012 to 2024. Solar Pro 4 93.2%
  • PRBench In use Model answers 1,100 expert-written law and finance questions. LLM judge grades against weighted expert rubrics of 10-30 criteria each. Muse Spark 1.3 61.6%