Benchmarks · agentic
Elimination Game Benchmark
Multi-player tournament where 8 LLM players converse, form alliances, and vote to eliminate each other, with final rankings scored via TrueSkill. Maintained by Lech Mazur. Score: TrueSkill mean (μ), with σ uncertainty reported separately.
48 models · 48 results · data as of 2026-10-29 · raw.githubusercontent.com ↗
Leaderboard
GPT-5.2 medium
All results
48 rows| # | Model | Effort | Score |
|---|---|---|---|
| 1 | GPT-5.2 | medium | 7.5 |
| 2 | GPT-5 | medium | 6.0 |
| 3 | GPT-5 Mini | medium | 5.7 |
| 4 | Opus 4.5 | high | 5.7 |
| 5 | Gemini 3 Flash | high | 5.7 |
| 6 | Grok 3 Mini | high | 5.5 |
| 7 | GPT-4o | high | 5.5 |
| 8 | DeepSeek R1 | xhigh | 5.3 |
| 9 | Opus 4.1 | high | 5.2 |
| 10 | Sonnet 4.5 | high | 5.2 |
| 11 | Grok 4 | high | 5.2 |
| 12 | GPT-4.5 | 5.1 | |
| 13 | Sonnet 2024.10 | 5.1 | |
| 14 | Grok 3 | none | 5.1 |
| 15 | Gemini 3 Pro | high | 4.9 |
| 16 | Sonnet 3.7 | medium | 4.8 |
| 17 | Gemini 2.5 Flash | medium | 4.7 |
| 18 | Sonnet 4 | high | 4.6 |
| 19 | MiniMax M2 | 4.6 | |
| 20 | Qwen 3 Max | max | 4.5 |
| 21 | o3 | medium | 4.5 |
| 22 | Gemini 2.5 Pro | high | 4.5 |
| 23 | Opus 4 | medium | 4.4 |
| 24 | Qwen 3 235B-A22B | xhigh | 4.4 |
| 25 | o3 mini | medium | 4.4 |
| 26 | Kimi K2 | medium | 4.3 |
| 27 | GLM 4.5 | xhigh | 4.2 |
| 28 | Mistral Large 2 | none | 4.1 |
| 29 | DeepSeek V3 | none | 4.1 |
| 30 | o1 | medium | 3.9 |