All Benchmarks/In-context learning

EBR-bench (Earthborne Rangers)

Higher is better
What does it measure?
A learning-on-the-job test by Epoch AI: the AI plays the obscure card board game Earthborne Rangers repeatedly to see whether it improves with practice.
Who evaluates it, and how?
The AI plays 10 times, learning and taking notes in the first 8, and the last 2 are scored, using the better one. Each play covers 5 in-game days and is scored by how many of 21 objectives are achieved.
Reading the score
Scores run from 0 to 100%. Higher is better.
Top score90.5%Claude Fable 5.2
Models tested28
Last updated2026.11.06
Model release date (newest on the right)
Best score so farHigher on the chart is better
  1. 1
    Anthropic
    Claude Fable 5.2
    2026.11.09 · max
    90.5%
  2. 2
    Google
    Gemini 4 Carbon
    2026.10.28 · high
    87.6%
  3. 3
    OpenAI
    GPT-6.1 Terra
    2026.10.26 · max
    78.1%
  4. 4
    OpenAI
    GPT-6 Astra
    2026.09.04 · max
    76.2%
  5. 5
    Anthropic
    Claude Opus 5.5
    2026.09.22 · max
    71.4%
  6. 6
    Google
    Gemini 4 Argon
    2026.09.30 · high
    64.8%
  7. 7
    Anthropic
    Claude Fable 5.1
    2026.09.01 · max
    57.1%
  8. 8
    OpenAI
    GPT-6.1 Sol
    2026.09.29 · max
    54.3%
  9. 9
    OpenAI
    GPT-6 Sol
    2026.09.22 · max
    53.3%
  10. 10
    Anthropic
    Claude Opus 5
    2026.07.24 · max
    45.7%

The "harness" is the agent program the AI used to carry out the task. The same model can score very differently depending on its harness and reasoning effort.

Where the scores come from

  • Independent measurement: scores run and published by an organization other than the test’s maker, using the same method.