HumanEval.org

Benchmarks

LiveBench — Instruction Following

last retrieved 2026-09-08 21:40 UTC

Third-party scores shown as context with provenance — our own arena ratings from blind human votes stay alongside every row and remain the primary signal.

ModelLiveBench — Instruction FollowingObservedSource
Gemini 3.8 FlashGoogle81.42026-09-07source ↗
Gemini 3.1 Pro PreviewGoogle79.12026-09-07source ↗
Claude Fable 5Anthropic75.82026-09-07source ↗
GPT-6 AstraOpenAI75.62026-09-07source ↗
Qwen3.8 Max (0902)Alibaba Qwen74.12026-09-07source ↗
Claude Fable 5.1Anthropic73.02026-09-07source ↗
Claude Opus 4.8Anthropic72.02026-09-07source ↗
Grok 4.6SpaceXAI71.92026-09-07source ↗
GPT-5.6 SolOpenAI71.82026-09-07source ↗
Kimi K3Moonshot AI71.42026-09-07source ↗
GPT-5.4OpenAI70.22026-09-07source ↗
GLM-5.3Z.ai69.32026-09-07source ↗
DeepSeek V4 Pro (0813)DeepSeek67.72026-09-07source ↗
Claude Opus 4.7Anthropic66.72026-09-07source ↗
GPT-5.6 TerraOpenAI64.62026-09-07source ↗
Claude Sonnet 5Anthropic63.92026-09-07source ↗
Claude Opus 5Anthropic63.82026-09-07source ↗
GPT-5.6 LunaOpenAI60.12026-09-07source ↗

— = not yet evaluated by HumanEval (no eligible arena votes). External values are the newest observation per model; hover a value for source notes and a rating for its 95% CI. Unit: score, higher is better.

Not enough overlap with our arena ratings for a correlation scatter yet — fewer than two scored models are rated.