Benchmark Almanac

Surface Evolver Bench

Claude Fable 5 95% leads since June 2026
Not in model cards

Agents write Surface Evolver datafiles that model liquid surfaces under physical constraints across 16 tasks. Deterministic checks give partial credit for matching hidden reference quantities.

Only one example task is public, and the scored tasks and grading scripts stay hidden.

Made by
Yann Henon
Released
June 2026
Kind
Research science
Level
Expert
Scored
Checked automatically
Answers
Code
Tasks
Public, plus a private test set
Strongest model tested
Claude Fable 5 #11 on the capability index
Last model added
September 2026
Scores refreshed
November 12th, 2026
26
Entries on its own leaderboard, 25 models we track

2 models share the top score of 95% among models released by September 2026

Best score per entry from the benchmark's own leaderboard, read on November 12th, 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.

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:

26 entries. Scroll sideways for the rest.

  1. 95.0 Claude Fable 5
  2. 95.0 Kimi K3
  3. 93.1 GPT-5.6 Sol
  4. 88.1 GPT-5.5
  5. 87.5 Claude Opus 4.8
  6. 83.8 GPT-5.6 Terra
  7. 76.9 Gemini 3.8 Flash
  8. 74.4 Grok 4.5
  9. 61.9 GPT-5.6 Luna
  10. 60.0 Claude Sonnet 5
  11. 58.1 Gemini 3.5 Flash
  12. 55.6 GLM-5.2
  13. 55.0 MiniMax-M3
  14. 52.5 Muse Spark 1.1
  15. 52.5 GLM-5.3-Flash
  16. 48.8 Kimi K2.7 Code
  17. 46.3 DeepSeek V4.1 Flash
  18. 45.0 Qwen 3.8 27B
  19. 44.4 Qwen 3.6 35B-A3B
  20. 40.0 DeepSeek-V4-Pro
  21. 30.6 Gemma 4 31B IT
  22. 28.1 mistral-large-4-0
  23. 26.9 Mistral Medium 3.5
  24. 25.0 gpt-oss-120b
  25. 15.6 Trinity Large Thinking
  26. 15.6 Laguna M.1
  1. 1 Claude Fable 5 high 95%
  2. 2 Kimi K3 95%
  3. 3 GPT-5.6 Sol xhigh 93.1%
  4. 4 GPT-5.5 high 88.1%
  5. 5 Claude Opus 4.8 high 87.5%
  6. 6 GPT-5.6 Terra xhigh 83.8%
  7. 7 Gemini 3.8 Flash high 76.9%
  8. 8 Grok 4.5 high 74.4%
  9. 9 GPT-5.6 Luna medium 61.9%
  10. 10 Claude Sonnet 5 medium 60%
  11. 11 Gemini 3.5 Flash medium 58.1%
  12. 12 GLM-5.2 high 55.6%
  13. 13 MiniMax-M3 55%
  14. 14 Muse Spark 1.1 high 52.5%
  15. 15 GLM-5.3-Flash max 52.5%
  16. 16 Kimi K2.7 Code 48.8%
  17. 17 DeepSeek V4.1 Flash high 46.3%
  18. 18 Qwen 3.8 27B xhigh 45%
  19. 19 Qwen 3.6 35B-A3B 44.4%
  20. 20 DeepSeek-V4-Pro high 40%
  21. 21 Gemma 4 31B IT none 30.6%
  22. 22 mistral-large-4-0 28.1%
  23. 23 Mistral Medium 3.5 high 26.9%
  24. 24 gpt-oss-120b 25%
  25. 25 Trinity Large Thinking 15.6%
  26. 26 Laguna M.1 15.6%

Source: Official Surface Evolver Bench leaderboard

#EntryScoreSettingDateSource
1 Claude Fable 5 95% high Jun 2026 Official leaderboard
2 Kimi K3 95% Jul 2026 Official leaderboard
3 GPT-5.6 Sol 93.1% xhigh Jul 2026 Official leaderboard
4 GPT-5.5 88.1% high Apr 2026 Official leaderboard
5 Claude Opus 4.8 87.5% high May 2026 Official leaderboard
6 GPT-5.6 Terra 83.8% xhigh Jul 2026 Official leaderboard
7 Gemini 3.8 Flash 76.9% high Sep 2026 Official leaderboard
8 Grok 4.5 74.4% high Jul 2026 Official leaderboard
9 GPT-5.6 Luna 61.9% medium Jul 2026 Official leaderboard
10 Claude Sonnet 5 60% medium Jun 2026 Official leaderboard
11 Gemini 3.5 Flash 58.1% medium May 2026 Official leaderboard
12 GLM-5.2 55.6% high Jun 2026 Official leaderboard
13 MiniMax-M3 55% Jun 2026 Official leaderboard
14 Muse Spark 1.1 52.5% high Jul 2026 Official leaderboard
15 GLM-5.3-Flash 52.5% max Aug 2026 Official leaderboard
16 Kimi K2.7 Code 48.8% Jun 2026 Official leaderboard
17 DeepSeek V4.1 Flash 46.3% high Sep 2026 Official leaderboard
18 Qwen 3.8 27B 45% xhigh Aug 2026 Official leaderboard
19 Qwen 3.6 35B-A3B 44.4% Apr 2026 Official leaderboard
20 DeepSeek-V4-Pro 40% high Apr 2026 Official leaderboard
21 Gemma 4 31B IT 30.6% none Apr 2026 Official leaderboard
22 mistral-large-4-0 28.1% Official leaderboard
23 Mistral Medium 3.5 26.9% high Apr 2026 Official leaderboard
24 gpt-oss-120b 25% Aug 2025 Official leaderboard
25 Trinity Large Thinking 15.6% Apr 2026 Official leaderboard
26 Laguna M.1 15.6% Apr 2026 Official leaderboard

Surface Evolver score versus cost

The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

0% 20% 40% 60% 80% 100% $0$0.5$1$1.5 Mean cost per task ($), lower is better Mean score Trinity Large Thinking: Mean score 15.6%, mean cost per task $0.09 Qwen 3.6 35B-A3B: Mean score 44.4%, mean cost per task $0.15 GPT-5.5 (high): Mean score 88.1%, mean cost per task $0.86 GPT-5.5 (medium): Mean score 81.3%, mean cost per task $0.6 DeepSeek-V4-Pro (high): Mean score 40%, mean cost per task $0.26 Laguna M.1: Mean score 15.6%, mean cost per task $0.03 Mistral Medium 3.5 (high): Mean score 26.9%, mean cost per task $0.4 Gemini 3.5 Flash (medium): Mean score 58.1%, mean cost per task $0.44 Claude Opus 4.8 (high): Mean score 87.5%, mean cost per task $1.37 Claude Opus 4.8 (none): Mean score 68.1%, mean cost per task $0.61 MiniMax-M3: Mean score 55%, mean cost per task $0.13 Claude Fable 5 (high): Mean score 95%, mean cost per task $1.22 Kimi K2.7 Code: Mean score 48.8%, mean cost per task $0.25 GLM-5.2 (high): Mean score 55.6%, mean cost per task $0.29 Claude Sonnet 5 (medium): Mean score 60%, mean cost per task $0.75 Grok 4.5 (high): Mean score 74.4%, mean cost per task $0.29 Muse Spark 1.1 (high): Mean score 52.5%, mean cost per task $0.16 Qwen 3.8 27B (xhigh): Mean score 45%, mean cost per task $0.57 DeepSeek V4.1 Flash (high): Mean score 46.3%, mean cost per task $0.13 mistral-large-4-0: Mean score 28.1%, mean cost per task $0.38. Not in our model list gpt-oss-120b: Mean score 25%, mean cost per task $0.006 Gemma 4 31B IT (none): Mean score 30.6%, mean cost per task $0.01 GPT-5.6 Luna (medium): Mean score 61.9%, mean cost per task $0.06 GPT-5.6 Sol (xhigh): Mean score 93.1%, mean cost per task $0.59 GPT-5.6 Terra (xhigh): Mean score 83.8%, mean cost per task $0.47 Kimi K3: Mean score 95%, mean cost per task $0.76 GLM-5.3-Flash (max): Mean score 52.5%, mean cost per task $0.02 Gemini 3.8 Flash (high): Mean score 76.9%, mean cost per task $0.16 Kimi K3GPT-5.6 Sol (xhigh)GPT-5.6 Terra (xhigh)Gemini 3.8 Flash (high)GPT-5.6 Luna (medium)GLM-5.3-Flash (max)Gemma 4 31B IT (none)gpt-oss-120b 0% 20% 40% 60% 80% 100% $0$0.5$1$1.5 Mean cost per task ($), lower is better Mean score Trinity Large Thinking: Mean score 15.6%, mean cost per task $0.09 Qwen 3.6 35B-A3B: Mean score 44.4%, mean cost per task $0.15 GPT-5.5 (high): Mean score 88.1%, mean cost per task $0.86 GPT-5.5 (medium): Mean score 81.3%, mean cost per task $0.6 DeepSeek-V4-Pro (high): Mean score 40%, mean cost per task $0.26 Laguna M.1: Mean score 15.6%, mean cost per task $0.03 Mistral Medium 3.5 (high): Mean score 26.9%, mean cost per task $0.4 Gemini 3.5 Flash (medium): Mean score 58.1%, mean cost per task $0.44 Claude Opus 4.8 (high): Mean score 87.5%, mean cost per task $1.37 Claude Opus 4.8 (none): Mean score 68.1%, mean cost per task $0.61 MiniMax-M3: Mean score 55%, mean cost per task $0.13 Claude Fable 5 (high): Mean score 95%, mean cost per task $1.22 Kimi K2.7 Code: Mean score 48.8%, mean cost per task $0.25 GLM-5.2 (high): Mean score 55.6%, mean cost per task $0.29 Claude Sonnet 5 (medium): Mean score 60%, mean cost per task $0.75 Grok 4.5 (high): Mean score 74.4%, mean cost per task $0.29 Muse Spark 1.1 (high): Mean score 52.5%, mean cost per task $0.16 Qwen 3.8 27B (xhigh): Mean score 45%, mean cost per task $0.57 DeepSeek V4.1 Flash (high): Mean score 46.3%, mean cost per task $0.13 mistral-large-4-0: Mean score 28.1%, mean cost per task $0.38. Not in our model list gpt-oss-120b: Mean score 25%, mean cost per task $0.006 Gemma 4 31B IT (none): Mean score 30.6%, mean cost per task $0.01 GPT-5.6 Luna (medium): Mean score 61.9%, mean cost per task $0.06 GPT-5.6 Sol (xhigh): Mean score 93.1%, mean cost per task $0.59 GPT-5.6 Terra (xhigh): Mean score 83.8%, mean cost per task $0.47 Kimi K3: Mean score 95%, mean cost per task $0.76 GLM-5.3-Flash (max): Mean score 52.5%, mean cost per task $0.02 Gemini 3.8 Flash (high): Mean score 76.9%, mean cost per task $0.16 Kimi K3GPT-5.6 Sol (xhigh)GPT-5.6 Terra (xhigh)Gemini 3.8 Flash (high)GPT-5.6 Luna (medium)GLM-5.3-Flash (max)Gemma 4 31B IT (none)gpt-oss-120b
ModelSettingMean scoreMean cost per task
Kimi K3 95% $0.76
Claude Fable 5 high 95% $1.22
GPT-5.6 Sol xhigh 93.1% $0.59
GPT-5.5 high 88.1% $0.86
Claude Opus 4.8 high 87.5% $1.37
GPT-5.6 Terra xhigh 83.8% $0.47
GPT-5.5 medium 81.3% $0.6
Gemini 3.8 Flash high 76.9% $0.16
Grok 4.5 high 74.4% $0.29
Claude Opus 4.8 none 68.1% $0.61
GPT-5.6 Luna medium 61.9% $0.06
Claude Sonnet 5 medium 60% $0.75
Gemini 3.5 Flash medium 58.1% $0.44
GLM-5.2 high 55.6% $0.29
MiniMax-M3 55% $0.13
GLM-5.3-Flash max 52.5% $0.02
Muse Spark 1.1 high 52.5% $0.16
Kimi K2.7 Code 48.8% $0.25
DeepSeek V4.1 Flash high 46.3% $0.13
Qwen 3.8 27B xhigh 45% $0.57
Qwen 3.6 35B-A3B 44.4% $0.15
DeepSeek-V4-Pro high 40% $0.26
Gemma 4 31B IT none 30.6% $0.01
mistral-large-4-0 28.1% $0.38
Mistral Medium 3.5 high 26.9% $0.4
gpt-oss-120b 25% $0.01
Laguna M.1 15.6% $0.03
Trinity Large Thinking 15.6% $0.09

The best score, 95%, costs $0.76 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 Google DeepMind Moonshot AI Z.ai Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 $10 Cost per task, log scale Kimi K3 GPT-5.6 Terra, xhigh Gemini 3.8 Flash, high GPT-5.6 Luna, medium GLM-5.3-Flash, max Gemma 4 31B IT, none gpt-oss-120b 0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 $10 Cost per task, log scale Kimi K3 GPT-5.6 Terra, xhigh

Source: Official Surface Evolver Bench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Fable 5 high 95% $1.22
Kimi K3 95% $0.76
GPT-5.6 Sol xhigh 93.1% $0.59
GPT-5.5 high 88.1% $0.86
Claude Opus 4.8 high 87.5% $1.37
GPT-5.6 Terra xhigh 83.8% $0.47
GPT-5.5 medium 81.3% $0.60
Gemini 3.8 Flash high 76.9% $0.16
Grok 4.5 high 74.4% $0.29
Claude Opus 4.8 none 68.1% $0.61
GPT-5.6 Luna medium 61.9% $0.06
Claude Sonnet 5 medium 60% $0.75
Gemini 3.5 Flash medium 58.1% $0.44
GLM-5.2 high 55.6% $0.29
MiniMax-M3 55% $0.13
Muse Spark 1.1 high 52.5% $0.16
GLM-5.3-Flash max 52.5% $0.02
Kimi K2.7 Code 48.8% $0.25
DeepSeek V4.1 Flash high 46.3% $0.13
Qwen 3.8 27B xhigh 45% $0.57
Qwen 3.6 35B-A3B 44.4% $0.15
DeepSeek-V4-Pro high 40% $0.26
Gemma 4 31B IT none 30.6% $0.01
Mistral Medium 3.5 high 26.9% $0.40
gpt-oss-120b 25% $0.006
Trinity Large Thinking 15.6% $0.09
Laguna M.1 15.6% $0.03

2 models have held number one on the official leaderboard since its release in June 2026

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

OpenAI Anthropic
Own leaderboard Passed 85% in Jun 2026, by GPT-5.5
Claude Fable 5 95%

Source: Official Surface Evolver Bench leaderboard

GPT-5.5 on the official leaderboard passed 85% in April 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 Near the ceiling (85%) Claude Fable 5 GPT-5.5 Qwen 3.6 35B-A3B Gemma 4 31B IT gpt-oss-120b 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Fable 5 GPT-5.5

Source: Official Surface Evolver Bench leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in Science and reasoning

All 92 in Science and reasoning
  • GPQA Diamond In use 198 graduate-level 'Google-proof' multiple-choice questions in biology, chemistry and physics. Gemini 4 Carbon 96.5%
  • Terminal-Bench-Science In use 70 scientific research-workflow tasks across five science areas, run in Harbor containers. Hidden tests give the resolution rate. GPT-6 Astra 68.1%
  • FrontierScience In use Models solve physics, chemistry, and biology problems in an Olympiad track graded by short answer and a Research track graded by 10-point rubrics. Intern-S2-397B 87%
  • BioMysteryBench In use Bioinformatics problems on anonymized real datasets that require real analysis with tools. Rubric-graded, with human-solvable and human-difficult splits. Claude Fable 5.2 83.7%
  • HiPhO In use Models solve theory problems from 2024 and 2025 physics olympiad exams, some with diagrams. Official marking schemes grade answers and steps, mapped to medal levels. Gemini 3 Pro 78.3%
  • LABBench2 In use 11 biology research sub-tasks (literature, figures, tables, patents, trials and more) with tool access. Macro-averaged accuracy. GPT-5.2 60.9%