Benchmarks / Reasoning & Math
LiveBench Reasoning
Contamination-limited monthly-refreshed benchmark; category mean over its tasks.
maintained by LiveBench · livebench.ai · 40 observations on record
Versions
1 version
| Version | Tasks |
|---|---|
| Release 2026-06-25 current livebench-reasoning-2026-06-25 | — |
Current leaderboard
Release 2026-06-25 · Accuracy · top 25 of 40 systems
| # | System | Accuracy % |
|---|---|---|
| 1 | GPT 5.6 Solmax effort OpenAI | 91.7% |
| 2 | GPT 5.6 Terramax effort OpenAI | 90.6% |
| 3 | GPT 5.6 Solxhigh effort OpenAI | 90.2% |
| 4 | Claude Opus 4.8xhigh effort Anthropic | 89.7% |
| 5 | GPT 5.5 (2026-04-22)high effort OpenAI | 89.7% |
| 6 | Claude Fable 5max effort Anthropic | 89.7% |
| 7 | GPT 5.5 (2026-04-22)xhigh effort OpenAI | 89.7% |
| 8 | Claude Opus 4.8max effort Anthropic | 89.2% |
| 9 | Claude Sonnet 5xhigh effort Anthropic | 88.7% |
| 10 | Claude Opus 4.6high effort Anthropic | 88.7% |
| 11 | Muse Spark 1.1high effort Meta AI | 88.4% |
| 12 | GPT 5.4 (2026-03-05)xhigh effort OpenAI | 88.1% |
| 13 | Muse Spark 1.1xhigh effort Meta AI | 87.7% |
| 14 | Claude Fable 5xhigh effort Anthropic | 87.7% |
| 15 | GPT 5.5 (2026-04-22) OpenAI | 87.3% |
| 16 | Claude Opus 4.7xhigh effort Anthropic | 87.2% |
| 17 | Grok 4.5 xAI | 87.2% |
| 18 | GPT 5.6 Lunamax effort OpenAI | 85.6% |
| 19 | GPT 5.6 Terraxhigh effort OpenAI | 84.9% |
| 20 | Claude Sonnet 4.6medium effort Anthropic | 84.8% |
| 21 | GPT 5.6 Lunaxhigh effort OpenAI | 84.7% |
| 22 | Gemini 3.1 Pro Previewhigh effort Google DeepMind | 84.0% |
| 23 | Qwen3 7 Max Alibaba (Qwen) | 83.3% |
| 24 | GPT 5.2 (2025-12-11)high effort OpenAI | 83.2% |
| 25 | Kimi K2 7 Code Moonshot | 82.8% |
One row per evaluated system — reasoning-effort variants rank separately.