GBAEval
Rank #11 of 51 · 14 models scored · top score 74.47
unknown
Strongest on discrimination, weakest on saturation headroom. Caveat: the public test set leaves contamination on the table.
Weighted composite. Each signal is computed from the source-backed leaderboard below.
Does it still separate models?
How far from ceiling / clustered at the top?
Public vs held-out; training-leak risk.
Apples-to-apples, or mixed / vendor-optimized?
How old is the benchmark?
| # | Model | Score | Evidence | Measured |
|---|---|---|---|---|
| 1 | Claude Fable 5 | 74.47 | unverified | 2026-06-09 |
| 2 | Claude Opus 4.8 | 70.9 | unverified | 2026-05-28 |
| 3 | GPT-5.5 | 53.22 | unverified | 2026-04-23 |
| 4 | Claude Sonnet 4.6 | 48.76 | unverified | 2026-02-17 |
| 5 | Claude Opus 4.6 | 44.12 | unverified | 2026-02-05 |
| 6 | Claude Opus 4.7 | 43.81 | unverified | 2026-04-16 |
| 7 | GPT-5.4 | 31.6 | unverified | 2026-03-05 |
| 8 | Gemini 3.5 Flash | 6.7 | unverified | 2026-05-19 |
| 9 | Kimi K2.6 | 0.86 | unverified | 2026-04-20 |
| 10 | Gemini 3.1 Pro | 0.85 | unverified | 2026-02-19 |
| 11 | Qwen3.7-Max | 0.41 | unverified | 2026-05-19 |
| 12 | GLM-5.2 | 0.02 | unverified | 2026-06-16 |
| 13 | MiniMax-M2.7 | 0 | unverified | 2026-03-18 |
| 14 | GLM-5.1 | 0 | unverified | 2026-04-07 |