# GPQA Diamond — AI model rankings

Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better.

59 tracked models have a published GPQA Diamond score. Higher is better. Scores come from published lab reports and benchmark sources; recorded sources appear beside the scores.

## Ranking

| Rank | Model | Developer | Score | Source | Released |
| --- | --- | --- | --- | --- | --- |
| 1 | GPT-6 Astra | OpenAI | 96% | Lab | Sep 3 2026 |
| 2 | GPT-5.4-Pro | OpenAI | 94.4% | Lab | Mar 5 2026 |
| 3 | Gemini 3.1 Pro | Google | 94.3% | Lab | Feb 19 2026 |
| 4 | Claude Opus 4.7 | Anthropic | 94.2% | Lab | Apr 16 2026 |
| 5 | GPT-5.5 | OpenAI | 93.6% | Lab | Apr 23 2026 |
| 6 | Kimi K3 | Moonshot AI | 93.5% | Lab | Jul 16 2026 |
| 7 | GPT-5.4 | OpenAI | 92.8% | Lab | Mar 5 2026 |
| 8 | GPT-5.2 | OpenAI | 92.4% | Lab | Dec 11 2025 |
| 8 | Qwen3.7-Max | Qwen | 92.4% | Lab | May 20 2026 |
| 10 | Gemini 3.0 Pro | Google | 91.9% | Lab | Nov 18 2025 |
| 11 | Qwen3.8-Flash-Next | Qwen | 91.7% | Lab | Aug 26 2026 |
| 12 | Claude Opus 4.6 | Anthropic | 91.3% | Lab | Feb 5 2026 |
| 13 | GLM-5.2 | Z.ai | 91.2% | Lab | Jun 16 2026 |
| 14 | DeepSeek-V4.1-Flash | DeepSeek | 90.9% | Lab | Sep 10 2026 |
| 15 | Kimi K2.6 | Moonshot AI | 90.5% | Lab | Apr 21 2026 |
| 16 | Gemini 3.0 Flash | Google | 90.4% | Lab | Dec 17 2025 |
| 17 | DeepSeek-V4-Pro | DeepSeek | 90.1% | Lab | Apr 24 2026 |
| 18 | Claude Sonnet 4.6 | Anthropic | 89.9% | Lab | Feb 17 2026 |
| 19 | Muse Spark | Meta | 89.5% | Lab | Apr 8 2026 |
| 20 | Qwen3.8-27B | Qwen | 89.2% | Lab | Aug 14 2026 |
| 21 | Qwen3.5 | Qwen | 88.4% | Lab | Feb 16 2026 |
| 22 | GPT-5.1 | OpenAI | 88.1% | Lab | Nov 12 2025 |
| 22 | DeepSeek-V4-Flash | DeepSeek | 88.1% | Lab | Apr 24 2026 |
| 24 | Kimi K2.5 | Moonshot AI | 87.6% | Lab | Jan 27 2026 |
| 25 | Claude Opus 4.5 | Anthropic | 87% | Lab | Nov 24 2025 |
| 26 | Nemotron 3 Ultra | NVIDIA | 86.7% | Lab | Jun 4 2026 |
| 27 | Gemini 2.5 Pro | Google | 86.4% | Lab | Mar 25 2025 |
| 28 | GLM-5.1 | Z.ai | 86.2% | Lab | Apr 7 2026 |
| 29 | GLM-5 | Z.ai | 86% | Lab | Feb 12 2026 |
| 29 | Qwen3.6 | Qwen | 86% | Lab | Apr 16 2026 |
| 31 | GLM-4.7 | Z.ai | 85.7% | Lab | Dec 22 2025 |
| 32 | Muse Glimmer | Meta | 83.5% | Lab | Aug 10 2026 |
| 33 | Claude Sonnet 4.5 | Anthropic | 83.4% | Lab | Sep 29 2025 |
| 34 | Claude Opus 4.1 | Anthropic | 80.9% | Lab | Aug 5 2025 |
| 35 | gpt-oss-120b | OpenAI | 80.1% | Lab | Aug 5 2025 |
| 36 | Claude Opus 4 | Anthropic | 79.6% | Lab | May 22 2025 |
| 37 | Nemotron 3 Super | NVIDIA | 79.2% | Lab | Mar 11 2026 |
| 38 | GLM-4.5 | Z.ai | 79.1% | Lab | Jul 28 2025 |
| 39 | Llama Nemotron Ultra 253B | NVIDIA | 76% | Lab | Apr 8 2025 |
| 40 | o1 | OpenAI | 75.7% | Lab | Dec 5 2024 |
| 41 | Claude Sonnet 4 | Anthropic | 75.4% | Lab | May 22 2025 |
| 41 | Nemotron 3.5 Lightning | NVIDIA | 75.4% | Lab | Aug 11 2026 |
| 43 | Kimi K2 | Moonshot AI | 75.1% | Lab | Jul 11 2025 |
| 43 | Kimi K2 (0905) | Moonshot AI | 75.1% | Lab | Sep 5 2025 |
| 45 | o1-preview | OpenAI | 73.3% | Lab | Sep 12 2024 |
| 46 | Claude Haiku 4.5 | Anthropic | 73% | Lab | Oct 15 2025 |
| 46 | Nemotron 3 Nano | NVIDIA | 73% | Lab | Dec 15 2025 |
| 48 | gpt-oss-20b | OpenAI | 71.5% | Lab | Aug 5 2025 |
| 49 | Claude 3.7 Sonnet | Anthropic | 68% | Lab | Feb 24 2025 |
| 50 | Nemotron Nano 2 | NVIDIA | 64% | Lab | Aug 18 2025 |
| 51 | o1-mini | OpenAI | 60% | Lab | Sep 12 2024 |
| 52 | Claude 3.5 Sonnet | Anthropic | 59.4% | Lab | Jun 20 2024 |
| 53 | Claude 3 Opus | Anthropic | 50.4% | Lab | Mar 4 2024 |
| 54 | GPT-4o | OpenAI | 49.9% | Lab | May 13 2024 |
| 55 | GPT-4 Turbo | OpenAI | 42.5% | Lab | Nov 6 2023 |
| 56 | Claude 3.5 Haiku | Anthropic | 41.6% | Lab | Oct 22 2024 |
| 57 | Claude 3 Sonnet | Anthropic | 40.4% | Lab | Mar 4 2024 |
| 58 | GPT-4o mini | OpenAI | 40.2% | Lab | Jul 18 2024 |
| 59 | Claude 3 Haiku | Anthropic | 33.3% | Lab | Mar 4 2024 |


---

Canonical page: https://aireleasetracker.com/benchmark/gpqa
Full dataset: https://aireleasetracker.com/llms-full.txt · JSON: https://aireleasetracker.com/models.json
Source: AI Release Tracker (https://aireleasetracker.com). Most benchmark scores come from lab launch material; gathered results identify the leaderboard that published them.