Benchmarks · reasoning
PLE-Bench
Uses the overall hidden-test rubric score, selecting each model's best harness on the native 0-1 scale. Score: mean rubric score on the hidden test set. Collected by Jake Boggs' benchmark dashboard.
18 models · 18 results · data as of 2026-10-29 · ple-bench.com ↗
Leaderboard
Opus 5.5 max
All results
18 rows| # | Model | Effort | PLE-Bench |
|---|---|---|---|
| 1 | Opus 5.5 | max | 48.6% |
| 2 | GPT-6 Astra | max | 45.8% |
| 3 | Sonnet 5.5 | max | 45.1% |
| 4 | Opus 5 | max | 42.6% |
| 5 | Gemini 4 Argon | max | 41.1% |
| 6 | Fable 5 | max | 34.5% |
| 7 | Grok 4.7 | max | 30.2% |
| 8 | Gemini 3.8 Flash | max | 28.2% |
| 9 | GLM 5.2 | max | 25.0% |
| 10 | DeepSeek V4.1 Flash | max | 22.5% |
| 11 | Gemini 3.1 Pro | high | 14.8% |
| 12 | Kimi K2 | medium | 6.2% |
| 13 | Qwen 3 Coder | medium | 6.2% |
| 14 | MiniMax M2 | 5.7% | |
| 15 | gpt-oss-120b | high | 4.6% |
| 16 | Gemma 4 26B | xhigh | 4.5% |
| 17 | Qwen 3 | high | 4.3% |
| 18 | Llama 4 Maverick | medium | 0.7% |