GPQA Diamond
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.
| Model | GPQA Diamond | Observed | Source |
|---|---|---|---|
| GPT-6 AstraOpenAI | 95.8%93.1% – 98.5% | 2026-08-30 | source ↗ |
| Gemini 3.1 Pro PreviewGoogle | 94.4%91.3% – 97.6% | 2026-08-06 | source ↗ |
| Claude Opus 5Anthropic | 93.9%91.0% – 96.8% | 2026-07-24 | source ↗ |
| GPT-5.6 SolOpenAI | 93.5%90.4% – 96.6% | 2026-07-09 | source ↗ |
| GPT-5.6 TerraOpenAI | 93.3%90.3% – 96.3% | 2026-07-09 | source ↗ |
| Grok 4.6SpaceXAI | 93.2%90.2% – 96.2% | 2026-08-14 | source ↗ |
| Kimi K3Moonshot AI | 93.1%90.2% – 96.0% | 2026-07-16 | source ↗ |
| Qwen3.8 Max (0902)Alibaba Qwen | 92.7%89.4% – 96.0% | 2026-08-04 | source ↗ |
| DeepSeek V4 Pro (0813)DeepSeek | 91.7%88.7% – 94.7% | 2026-08-18 | source ↗ |
| GPT-5.6 LunaOpenAI | 91.6%88.2% – 95.0% | 2026-07-09 | source ↗ |
| Claude Opus 4.8Anthropic | 91.0%87.3% – 94.8% | 2026-06-07 | source ↗ |
| GLM-5.3Z.ai | 90.9%87.8% – 94.0% | 2026-08-24 | source ↗ |
| Claude Opus 4.7Anthropic | 86.4%81.6% – 91.2% | 2026-08-06 | source ↗ |
| Claude Fable 5Anthropic | 85.9%81.0% – 90.7% | 2026-08-06 | source ↗ |
| Claude Sonnet 5Anthropic | 80.3%74.8% – 85.9% | 2026-08-06 | source ↗ |
| Llama 4 MaverickMeta | 67.0%61.5% – 72.5% | 2025-04-08 | source ↗ |
— = 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: %, higher is better.
Not enough overlap with our arena ratings for a correlation scatter yet — fewer than two scored models are rated.