LLM Debate Benchmark
Chat & Writing · 2026-03-22
LLM debate benchmark in which models debate both sides of contentious topics and are rated through pairwise outcomes. The leaderboard reports Bradley-Terry ratings.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5.1 | 1738 |
| Claude Fable 5 | 1727 |
| Claude Opus 5 | 1724 |
| Claude Opus 5.5 | 1701 |
| Kimi K3 | 1698 |
| Opus 4.7 | 1649 |
| GPT-5.6 Sol | 1645 |
| GLM-5.3 | 1639 |
| Muse Spark 1.1 | 1638 |
| Opus 4.8 | 1631 |
| Grok 4.7 | 1620 |
| MiMo-V2.6-Pro | 1609 |
| GPT-6 Astra | 1606 |
| Grok 4.6 | 1604 |
| Muse Spark 1.3 | 1593 |
| Sonnet 5 | 1585 |
| DeepSeek V4.1 Flash | 1582 |
| Hy4 Preview | 1579 |
| DeepSeek V4 Pro 0813 | 1572 |
| GLM-5.2 | 1563 |
| Sonnet 4.6 | 1561 |
| GPT-5.4 | 1548 |
| GPT-5.5 | 1528 |
| GLM-5.1 | 1519 |
| Gemini 3.8 Flash | 1514 |
| Kimi K2.6 | 1511 |
| MiniMax M3 | 1502 |
| Gemini 3.1 Pro Preview | 1491 |
| Grok 4.5 | 1490 |
| MiMo-V2.5-Pro | 1484 |
| Qwen3.6-Max-Preview | 1475 |
| Gemini 3.7 Flash | 1455 |
| Kimi K2.5 | 1454 |
| Doubao Seed 2.0 Pro | 1454 |
| Qwen3.7-Max (source unspecified) | 1434 |
| MiniMax M2.7 | 1434 |
| Grok 4.20 | 1412 |
| Gemini 3.5 Flash | 1407 |
| MiMo-V2-Pro | 1392 |
| Qwen3.5 397B A17B | 1390 |