Benchmarks / Language
LiveBench Language
Contamination-limited monthly-refreshed benchmark; category mean over its tasks.
maintained by LiveBench · livebench.ai · 36 observations on record
Versions
1 version
| Version | Tasks |
|---|---|
| Release 2026-06-25 current livebench-language-2026-06-25 | — |
Current leaderboard
Release 2026-06-25 · Accuracy · top 25 of 36 systems
| # | System | Accuracy % |
|---|---|---|
| 1 | Claude Fable 5max effort Anthropic | 90.7% |
| 2 | Claude Fable 5xhigh effort Anthropic | 89.5% |
| 3 | GPT 5.5 (2026-04-22)high effort OpenAI | 87.8% |
| 4 | GPT 5.6 Solmax effort OpenAI | 87.7% |
| 5 | GPT 5.5 (2026-04-22)xhigh effort OpenAI | 87.4% |
| 6 | GPT 5.6 Solxhigh effort OpenAI | 85.9% |
| 7 | GPT 5.5 (2026-04-22) OpenAI | 85.6% |
| 8 | Gemini 3.1 Pro Previewhigh effort Google DeepMind | 85.4% |
| 9 | Gemini 3.5 Flashhigh effort Google DeepMind | 84.6% |
| 10 | Claude Opus 4.6high effort Anthropic | 83.3% |
| 11 | GPT 5.6 Terramax effort OpenAI | 82.9% |
| 12 | Grok 4.5 xAI | 82.8% |
| 13 | GPT 5.4 (2026-03-05)xhigh effort OpenAI | 82.6% |
| 14 | Claude Opus 4.8xhigh effort Anthropic | 81.4% |
| 15 | Claude Opus 4.5 20251101high effort · 64K budget Anthropic | 81.3% |
| 16 | GPT 5.6 Terraxhigh effort OpenAI | 79.9% |
| 17 | GPT 5.2 (2025-12-11)high effort OpenAI | 79.8% |
| 18 | Qwen3 7 Max Alibaba (Qwen) | 79.7% |
| 19 | Deepseek V4 Pro DeepSeek | 78.1% |
| 20 | Kimi K2 7 Code Moonshot | 77.9% |
| 21 | Claude Opus 4.7xhigh effort Anthropic | 77.9% |
| 22 | Minimax M3 MiniMax | 76.8% |
| 23 | GLM 5.2 Z.ai | 76.2% |
| 24 | Claude Sonnet 4.6medium effort Anthropic | 76.1% |
| 25 | Kimi K2 6thinking Moonshot AI | 75.1% |
One row per evaluated system — reasoning-effort variants rank separately.