Benchmark
GPQA Diamond
Science benchmarks measure how well models answer science questions. Scores are percentages. Higher is better.
- Models and systems measured
- 282
- Scores recorded
- 1623
- Sources
- 4
- IQ weight
- 1.72
Score distribution
- 79.3%
- Mean
- 84.5%
- Median
- 15.7%
- Std dev
- 56.8%
- 10th percentile
- 92.2%
- 90th percentile
- 0.168%
- Min
This benchmark has a weight of 1.72 for chat models, measured over 272 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model or system | |
|---|---|---|
| 1 | GPT-6.2 Astramax· Epoch AI Benchmarking Hub data | 96.8% |
| 2 | Gemini 4 Hydrogenxhigh· Epoch AI Benchmarking Hub data | 96.8% |
| 3 | Gemini 4 Carbonhigh· Epoch AI Benchmarking Hub data | 96.5% |
| 4 | Gemini 4 ArgonOpenRouter harness, google/gemini-4-argon-20261013, Google, public 90-day aggregate· OpenRouter benchmark exports | 96.5% |
| 5 | Gemini 4 Argonhigh· Epoch AI Benchmarking Hub data | 96.4% |
| 6 | Gemini 4 Neonxhigh· Epoch AI Benchmarking Hub data | 96.3% |
| 7 | GPT-6 Astrahighest effort· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 96% |
| 8 | GPT-6 Astramax· Epoch AI Benchmarking Hub data | 95.8% |
| 9 | Gemini 3.8 FlashOpenRouter harness, google/gemini-3.8-flash-20260902, auto-routing, public 90-day aggregate· OpenRouter benchmark exports | 95.6% |
| 10 | Claude Sonnet 5.5max· Epoch AI Benchmarking Hub data | 95.6% |
| 11 | Fugu UltraFugu official headline evaluation· Reported by Sakana AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 95.5% |
| 12 | GPT-6 AstraOpenRouter harness, openai/gpt-6-astra-pro-20260903, auto-routing, public 90-day aggregate, pro· OpenRouter benchmark exports | 95.5% |
| 13 | Gemini 3.8 Flashhigh· Epoch AI Benchmarking Hub data | 95.4% |
| 14 | GPT-6.1 Solmax· Epoch AI Benchmarking Hub data | 95.4% |
| 15 | Gemini 3.8 FlashOpenAI launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 95.3% |
| 16 | Gemini 3.8 FlashOpenRouter harness, google/gemini-3.8-flash-20260902, Google AI Studio, public 90-day aggregate· OpenRouter benchmark exports | 95.3% |
| 17 | Gemini 3.1 Pro PreviewOpenRouter harness, google/gemini-3.1-pro-preview-20260219, Google, public 90-day aggregate· OpenRouter benchmark exports | 95.2% |
| 18 | GPT-6 AstraOpenRouter harness, openai/gpt-6-astra-pro-20260903, OpenAI, public 90-day aggregate, pro· OpenRouter benchmark exports | 95.1% |
| 19 | Claude Fable 5.2OpenRouter harness, anthropic/claude-fable-5.2-20261109, Anthropic, public 90-day aggregate· OpenRouter benchmark exports | 95.1% |
| 20 | Gemini 3.1 Pro PreviewOpenRouter harness, google/gemini-3.1-pro-preview-20260219, Google AI Studio, public 90-day aggregate· OpenRouter benchmark exports | 95.1% |
| 21 | Claude Fable 5.2OpenRouter harness, anthropic/claude-fable-5.2-20261109, Azure, public 90-day aggregate· OpenRouter benchmark exports | 94.9% |
| 22 | GPT-6 AstraOpenRouter harness, openai/gpt-6-astra-pro-20260903, Azure, public 90-day aggregate, pro· OpenRouter benchmark exports | 94.9% |
| 23 | GPT-6 Astralower-cost setting· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 94.9% |
| 24 | Gemini 3.8 FlashOpenRouter harness, google/gemini-3.8-flash-20260902, Google, public 90-day aggregate· OpenRouter benchmark exports | 94.8% |
| 25 | Gemini 3.7 Flashhigh· Epoch AI Benchmarking Hub data | 94.8% |
| 26 | Ling 3.1 FlashOpenRouter harness, inclusionai/ling-3.1-flash-20261002, DeepInfra, public 90-day aggregate· OpenRouter benchmark exports | 94.8% |
| 27 | GPT-6.1 SolOpenRouter harness, openai/gpt-6.1-sol-20260929, Amazon Bedrock, public 90-day aggregate· OpenRouter benchmark exports | 94.8% |
| 28 | GPT-6 AstraOpenRouter harness, openai/gpt-6-astra-20260903, OpenAI, public 90-day aggregate· OpenRouter benchmark exports | 94.7% |
| 29 | Claude Fable 5.2OpenRouter harness, anthropic/claude-fable-5.2-20261109, Google, public 90-day aggregate· OpenRouter benchmark exports | 94.6% |
| 30 | GPT-6 AstraOpenRouter harness, openai/gpt-6-astra-20260903, Amazon Bedrock, public 90-day aggregate· OpenRouter benchmark exports | 94.6% |
| 31 | GPT-5.4 Proxhigh· Epoch AI Benchmarking Hub data | 94.6% |
| 32 | GPT-5.6 SolOpenAI GPT-5.6 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 94.6% |
| 33 | GPT-5.6 SolOpenAI launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 94.6% |
| 34 | Gemini 3.1 Pro PreviewOpenRouter harness, google/gemini-3.1-pro-preview-20260219, auto-routing, public 90-day aggregate· OpenRouter benchmark exports | 94.5% |
| 35 | Gemini 3.7 FlashOpenRouter harness, google/gemini-3.7-flash-20260813, Google, public 90-day aggregate· OpenRouter benchmark exports | 94.5% |
| 36 | Gemini 3.1 Pro Previewhigh· Epoch AI Benchmarking Hub data | 94.4% |
| 37 | GPT-6 AstraOpenRouter harness, openai/gpt-6-astra-20260903, auto-routing, public 90-day aggregate· OpenRouter benchmark exports | 94.4% |
| 38 | GPT-6.1 SolOpenRouter harness, openai/gpt-6.1-sol-20260929, auto-routing, public 90-day aggregate· OpenRouter benchmark exports | 94.4% |
| 39 | GPT-6.1 SolOpenRouter harness, openai/gpt-6.1-sol-20260929, OpenAI, public 90-day aggregate· OpenRouter benchmark exports | 94.4% |
| 40 | GPT-5.4 ProOpenAI GPT-5.4 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 94.4% |
| 41 | GPT-6 Solmax· Epoch AI Benchmarking Hub data | 94.3% |
| 42 | GPT-6.1 Terramax· Epoch AI Benchmarking Hub data | 94.3% |
| 43 | GPT-6 AstraOpenRouter harness, openai/gpt-6-astra-20260903, Azure, public 90-day aggregate· OpenRouter benchmark exports | 94.3% |
| 44 | Gemini 3.1 Pro Previewno tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 94.3% |
| 45 | Gemini 3.1 Pro PreviewOpenAI GPT-5.6 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 94.3% |
| 46 | Gemini 3.1 Pro PreviewOpus 4.8 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 94.3% |
| 47 | Claude Opus 4.7198 questions · adaptive thinking · max effort · avg@10· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 94.2% |
| 48 | Claude Opus 4.7Opus 4.8 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 94.2% |
| 49 | Gemini 3.6 Flashhigh· Epoch AI Benchmarking Hub data | 94.1% |
| 50 | Gemini 3.7 FlashOpenRouter harness, google/gemini-3.7-flash-20260813, Google AI Studio, public 90-day aggregate· OpenRouter benchmark exports | 94.1% |
Showing 1–50 of 1,621 results.
Not ranked
The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.
| Model or system | |
|---|---|
| MiMo V2 Flash Baseofficial Base table · 5-shot· Reported by Xiaomi MiMo· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 55.1% |
| Qwen3.8-2.4T-A95Blifecycle-page core-benchmark claim· Reported by Alibaba Cloud Model Studio· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 92.6% |
Tokens the model generated per task, as measured by the source. Comparable within this benchmark only.
Data credits: Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · OpenRouter public benchmark leaderboard and search-lane exports, CC BY 4.0. Fractional source scores converted to percentages; immutable dated snapshots. (licence) · OpenEvals/leaderboard-data, MIT (licence)