Benchmark Almanac

EBR-bench

Claude Fable 5.2 90.5% leads since November 2026
Not in model cards

Agent plays the board game Earthborne Rangers 10-30 times in a row. Scored on objectives completed (of 21) to test learning from experience.

One card was banned in 2026-10 after it broke the game. Epoch expects saturation soon. Game material is not distributed.

Made by
Epoch AI
Released
July 2026
Kind
Puzzles and chess
Level
Everyday
Scored
Checked automatically
Answers
Agent task
Tasks
Kept private
Strongest model tested
Claude Fable 5.2 #1 on the capability index
Last model added
November 2026
Scores refreshed
November 12th, 2026
31
Models scored in Epoch AI’s runs

Claude Fable 5.2 leads with 90.5%

Best score per model from Epoch AI’s own runs. Where a model ran with several settings or harnesses, we keep its best.

31 models. Scroll sideways for the rest.

  1. 90.5 Claude Fable 5.2
  2. 87.6 Gemini 4 Carbon
  3. 78.1 GPT-6.1 Terra
  4. 76.2 GPT-6 Astra
  5. 71.4 Claude Opus 5.5
  6. 64.8 Gemini 4 Argon
  7. 57.1 Claude Fable 5.1
  8. 54.3 GPT-6.1 Sol
  9. 53.3 GPT-6 Sol
  10. 45.7 Claude Opus 5
  11. 44.8 GPT-5.6 Sol
  12. 43.8 Claude Haiku 5.5
  13. 39.5 Claude Fable 5
  14. 34.3 GPT-5.5
  15. 31.4 Muse Wonder
  16. 30.5 Grok 4.6
  17. 28.6 Claude Opus 4.8
  18. 25.4 GPT-5.4
  19. 23.0 GPT-5.2
  20. 19.1 Claude Opus 4.7
  21. 18.1 Mistral Large 4
  22. 14.3 Claude Opus 4.5
  23. 14.3 Gemini 3.1 Pro
  24. 12.7 GPT-5
  25. 12.7 Claude Opus 4.6
  26. 9.52 Qwen3.7-Max
  27. 9.52 GLM-5.2
  28. 7.94 Claude Opus 4.1
  29. 4.76 Gemini 3.5 Flash
  30. 2.38 Claude Sonnet 4.5
  31. 2.38 Kimi K2.6
  1. 1 Claude Fable 5.2 max 90.5%
  2. 2 Gemini 4 Carbon high 87.6%
  3. 3 GPT-6.1 Terra max 78.1%
  4. 4 GPT-6 Astra max 76.2%
  5. 5 Claude Opus 5.5 max 71.4%
  6. 6 Gemini 4 Argon high 64.8%
  7. 7 Claude Fable 5.1 max 57.1%
  8. 8 GPT-6.1 Sol max 54.3%
  9. 9 GPT-6 Sol max 53.3%
  10. 10 Claude Opus 5 max 45.7%
  11. 11 GPT-5.6 Sol max 44.8%
  12. 12 Claude Haiku 5.5 max 43.8%
  13. 13 Claude Fable 5 max 39.5%
  14. 14 GPT-5.5 xhigh 34.3%
  15. 15 Muse Wonder max 31.4%
  16. 16 Grok 4.6 xhigh 30.5%
  17. 17 Claude Opus 4.8 max 28.6%
  18. 18 GPT-5.4 xhigh 25.4%
  19. 19 GPT-5.2 xhigh 23%
  20. 20 Claude Opus 4.7 max 19.1%
  21. 21 Mistral Large 4 high 18.1%
  22. 22 Claude Opus 4.5 128k thinking 14.3%
  23. 23 Gemini 3.1 Pro 14.3%
  24. 24 GPT-5 high 12.7%
  25. 25 Claude Opus 4.6 max 12.7%
  26. 26 Qwen3.7-Max 9.52%
  27. 27 GLM-5.2 max 9.52%
  28. 28 Claude Opus 4.1 7.94%
  29. 29 Gemini 3.5 Flash high 4.76%
  30. 30 Claude Sonnet 4.5 no thinking 2.38%
  31. 31 Kimi K2.6 2.38%

Source: Epoch AI runs (CC BY 4.0)

#ModelScoreSettingDateSource
1 Claude Fable 5.2 90.5% max Nov 2026 Epoch AI run
2 Gemini 4 Carbon 87.6% high Oct 2026 Epoch AI run
3 GPT-6.1 Terra 78.1% max Oct 2026 Epoch AI run
4 GPT-6 Astra 76.2% max Sep 2026 Epoch AI run
5 Claude Opus 5.5 71.4% max Sep 2026 Epoch AI run
6 Gemini 4 Argon 64.8% high Sep 2026 Epoch AI run
7 Claude Fable 5.1 57.1% max Sep 2026 Epoch AI run
8 GPT-6.1 Sol 54.3% max Sep 2026 Epoch AI run
9 GPT-6 Sol 53.3% max Sep 2026 Epoch AI run
10 Claude Opus 5 45.7% max Jul 2026 Epoch AI run
11 GPT-5.6 Sol 44.8% max Jul 2026 Epoch AI run
12 Claude Haiku 5.5 43.8% max Oct 2026 Epoch AI run
13 Claude Fable 5 39.5% max Jun 2026 Epoch AI run
14 GPT-5.5 34.3% xhigh Apr 2026 Epoch AI run
15 Muse Wonder 31.4% max Oct 2026 Epoch AI run
16 Grok 4.6 30.5% xhigh Aug 2026 Epoch AI run
17 Claude Opus 4.8 28.6% max May 2026 Epoch AI run
18 GPT-5.4 25.4% xhigh Mar 2026 Epoch AI run
19 GPT-5.2 23% xhigh Dec 2025 Epoch AI run
20 Claude Opus 4.7 19.1% max Apr 2026 Epoch AI run
21 Mistral Large 4 18.1% high Oct 2026 Epoch AI run
22 Claude Opus 4.5 14.3% 128k thinking Nov 2025 Epoch AI run
23 Gemini 3.1 Pro 14.3% Feb 2026 Epoch AI run
24 GPT-5 12.7% high Aug 2025 Epoch AI run
25 Claude Opus 4.6 12.7% max Feb 2026 Epoch AI run
26 Qwen3.7-Max 9.52% May 2026 Epoch AI run
27 GLM-5.2 9.52% max Jun 2026 Epoch AI run
28 Claude Opus 4.1 7.94% Aug 2025 Epoch AI run
29 Gemini 3.5 Flash 4.76% high May 2026 Epoch AI run
30 Claude Sonnet 4.5 2.38% no thinking Sep 2025 Epoch AI run
31 Kimi K2.6 2.38% Apr 2026 Epoch AI run

8 models have held number one in Epoch AI’s runs since its release in July 2026

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

Anthropic OpenAI Google DeepMind
Epoch AI runs Passed 85% in Oct 2026, by Gemini 4 Carbon
Claude Fable 5.2 90.5%

Source: Epoch AI runs (CC BY 4.0)

Epoch AI’s run of Gemini 4 Carbon passed 85% in October 2026

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

Run by Epoch AI Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Claude Fable 5.2 GPT-6.1 Terra Claude Fable 5.1 Claude Opus 5 Claude Fable 5 GPT-5.5 GPT-5.4 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Claude Fable 5.2 GPT-6.1 Terra

Source: Epoch AI runs (CC BY 4.0)

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 Games and long tasks

All 82 in Games and long tasks
  • ARC-AGI-3 In use Interactive game environments where agents must explore, infer goals and rules, and act efficiently without instructions. GPT-6 Astra 99.95%
  • Orak Fading LLM agents play 12 popular video games across most genres through an MCP interface. Each game scores in-game progress or outcomes, and the board ranks models by their average rank across the games. Gemini 2.5 Pro 3.90
  • MineDojo Fading Agents complete open-ended tasks in Minecraft, such as mining blocks and fighting mobs, from a large simulation suite. Programmatic checks score many tasks. Seed2.0 Pro 49%
  • TextQuests Only in system cards Agents play 25 classic interactive fiction games. The score is harmful actions per 100 steps, where lower is better. The board also reports game progress. Llama 4 Scout 7.00
  • GameWorld Only in system cards Multimodal game agents play 34 browser games across 170 tasks. Outcomes are checked from the game state. GPT-6.1 Terra 76.8%
  • Swarm Bench Only in system cards Models orchestrate agent swarms on complex tasks that benefit from coordinated decomposition and parallel execution. Kimi K3 76.3%