evals.report
BenchmarksLabsCompareRun guidesIn the wild
BenchmarksReasoning

LiveBench

A frequently updated public benchmark suite spanning reasoning, coding, math, language, and instruction-following tasks.

ReasoningscoreHigher is better
ScoresAboutRun this benchmark

What is LiveBench?

A frequently updated public benchmark suite spanning reasoning, coding, math, language, and instruction-following tasks. evals.report tracks reported LiveBench scores with the model, source, status, date, and run caveats attached — official leaderboard scores, vendor-reported launches, and clearly labeled community runs.

Top reported LiveBench score: GPT-5.5 — 80.71% (score).

ModelLabScore↓Source modelStatusDate
GPT-5.5OpenAI80.71%gpt-5.5-xhighOfficialApr 23, 2026Details
GPT-5.4OpenAI80.28%gpt-5.4-xhighOfficialMar 5, 2026Details
Gemini 3.1 Pro PreviewGoogle DeepMind79.93%gemini-3.1-pro-preview-highOfficialFeb 19, 2026Details
Claude Opus 4.8Anthropic77.22%claude-opus-4-8-xhigh-effortOfficialMay 28, 2026Details
Claude Opus 4.7Anthropic76.91%claude-opus-4-7-xhigh-effortOfficialApr 16, 2026Details
Claude Opus 4.6Anthropic76.33%claude-opus-4-6-thinking-auto-high-effortOfficialFeb 5, 2026Details
Claude Opus 4.5Anthropic75.96%claude-opus-4-5-20251101-thinking-64k-high-effortOfficialNov 24, 2025Details
Claude Sonnet 4.6Anthropic75.47%claude-sonnet-4-6-thinking-auto-medium-effortOfficialFeb 17, 2026Details
Gemini 3.5 FlashGoogle DeepMind75.02%gemini-3.5-flash-highOfficialMay 19, 2026Details
GPT-5.2OpenAI74.84%gpt-5.2-2025-12-11-highOfficialDec 11, 2025Details
Qwen3.7 Max PreviewAlibaba / Qwen74.29%qwen3.7-maxOfficialMay 14, 2026Details
DeepSeek V4 ProOpenDeepSeek73.58%deepseek-v4-proOfficialApr 24, 2026Details
Gemini 3 ProGoogle DeepMind73.39%gemini-3-pro-preview-11-2025-highOfficialNov 18, 2025Details
Kimi K2.6OpenMoonshot AI72.17%kimi-k2.6-thinkingOfficialApr 20, 2026Details
GLM-5.1OpenZ.ai70.18%glm-5.1OfficialApr 7, 2026Details
Grok 4.20 beta reasoningxAI67.96%grok-4.20-beta-0309-reasoningOfficialMar 9, 2026Details

Each row reports the model’s score on LiveBench. Click a row for the full run context.