Skip to content

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
0.168%96.8%
The band covers the middle 80% of scores. The line marks the median.

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
901DeepSeek Reasoner· Epoch AI Benchmarking Hub data83.4%
902Gemma 4 31B ITOpenRouter harness, google/gemma-4-31b-it-20260402, DeepInfra, public 90-day aggregate· OpenRouter benchmark exports83.4%
903Inkling FP4OpenRouter harness, thinkingmachines/inkling-20260715, auto-routing, public 90-day aggregate· OpenRouter benchmark exports83.4%
904GLM 5.3OpenRouter harness, z-ai/glm-5.3-20260816, Cohere, public 90-day aggregate· OpenRouter benchmark exports83.4%
905GLM 4.7OpenRouter harness, z-ai/glm-4.7-20251222, auto-routing, public 90-day aggregate· OpenRouter benchmark exports83.4%
906GLM 5.2OpenRouter harness, z-ai/glm-5.2-20260616, Friendli, public 90-day aggregate· OpenRouter benchmark exports83.4%
907GLM 5.3OpenRouter harness, z-ai/glm-5.3-20260816, Alibaba, public 90-day aggregate· OpenRouter benchmark exports83.4%
908Qwen 3.8 27BOpenRouter harness, qwen/qwen3.8-27b-20260814, DekaLLM, public 90-day aggregate· OpenRouter benchmark exports83.4%
909Claude Sonnet 4.5launch-page setting· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.83.4%
910Claude Sonnet 4.5Opus 4.6 System Card capability summary · adaptive thinking · max effort · default sampling · 5 trials unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.83.4%
911DeepSeek V4 Flash Vision ExpOpenRouter harness, deepseek/deepseek-v4-flash-vision-exp-20260821, AtlasCloud, public 90-day aggregate· OpenRouter benchmark exports83.4%
912GLM 5.2OpenRouter harness, z-ai/glm-5.2-20260616, StreamLake, public 90-day aggregate· OpenRouter benchmark exports83.3%
913Claude Fable 5high· Epoch AI Benchmarking Hub data83.3%
914Claude Sonnet 4.6high· Epoch AI Benchmarking Hub data83.3%
915Claude Sonnet 4.6medium· Epoch AI Benchmarking Hub data83.3%
916Gemini 3.5 Flash Litehigh· Epoch AI Benchmarking Hub data83.3%
917Qwen3.6 Flash· Epoch AI Benchmarking Hub data83.3%
918Qwen 3.8 27BOpenRouter harness, qwen/qwen3.8-27b-20260814, DeepInfra, public 90-day aggregate· OpenRouter benchmark exports83.3%
919Qwen3.8 FlashOpenRouter harness, qwen/qwen3.8-flash-20260826, Makora, public 90-day aggregate· OpenRouter benchmark exports83.3%
920GLM 5.3 FlashOpenRouter harness, z-ai/glm-5.3-flash-20260826, Cloudflare, public 90-day aggregate· OpenRouter benchmark exports83.3%
921Claude Opus 4extended thinking · parallel test-time compute· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.83.3%
922GLM 5.1OpenRouter harness, z-ai/glm-5.1-20260406, Friendli, public 90-day aggregate· OpenRouter benchmark exports83.3%
923GLM 5.3 FlashOpenRouter harness, z-ai/glm-5.3-flash-20260826, Relace, public 90-day aggregate· OpenRouter benchmark exports83.3%
924GLM 5.3 FlashOpenRouter harness, z-ai/glm-5.3-flash-20260826, SiliconFlow, public 90-day aggregate· OpenRouter benchmark exports83.3%
925Qwen 3.8 27BOpenRouter harness, qwen/qwen3.8-27b-20260814, SiliconFlow, public 90-day aggregate· OpenRouter benchmark exports83.2%
926Qwen 3.8 27BOpenRouter harness, qwen/qwen3.8-27b-20260814, Alibaba, public 90-day aggregate· OpenRouter benchmark exports83.2%
927DeepSeek V3.2OpenRouter harness, deepseek/deepseek-v3.2-20251201, AtlasCloud, public 90-day aggregate· OpenRouter benchmark exports83.2%
928Gemini 3 Flash· Epoch AI Benchmarking Hub data83.2%
929GLM 5.3 FlashOpenRouter harness, z-ai/glm-5.3-flash-20260826, BaseTen, public 90-day aggregate· OpenRouter benchmark exports83.2%
930GLM 5.3OpenRouter harness, z-ai/glm-5.3-20260816, Parasail, public 90-day aggregate· OpenRouter benchmark exports83.2%
931DeepSeek V3.1OpenRouter harness, deepseek/deepseek-chat-v3.1, Novita, public 90-day aggregate· OpenRouter benchmark exports83.2%
932Hy 4OpenRouter harness, tencent/hy4-20261104, Tencent, public 90-day aggregate· OpenRouter benchmark exports83.2%
933MiniMax M2.5OpenRouter harness, minimax/minimax-m2.5-20260211, Friendli, public 90-day aggregate· OpenRouter benchmark exports83.2%
934Qwen 3.8 27BOpenRouter harness, qwen/qwen3.8-27b-20260814, Cerebras, public 90-day aggregate· OpenRouter benchmark exports83.2%
935Qwen 3.8 27BOpenRouter harness, qwen/qwen3.8-27b-20260814, PrimeIntellect, public 90-day aggregate· OpenRouter benchmark exports83.2%
936GLM 5.3 FlashOpenRouter harness, z-ai/glm-5.3-flash-20260826, Crusoe, public 90-day aggregate· OpenRouter benchmark exports83.1%
937Qwen 3.8 27BOpenRouter harness, qwen/qwen3.8-27b-20260814, Parasail, public 90-day aggregate· OpenRouter benchmark exports83.1%
938GLM 5.3OpenRouter harness, z-ai/glm-5.3-20260816, Amazon Bedrock, public 90-day aggregate· OpenRouter benchmark exports83.1%
939Step 3.5 FlashSenseNova 6.7 Flash-Lite official benchmark chart · chart setting· Reported by SenseTime / OpenSenseNova· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.83.1%
940Qwen 3.8 27BOpenRouter harness, qwen/qwen3.8-27b-20260814, Wafer, public 90-day aggregate· OpenRouter benchmark exports83.1%
941GLM 5.3 FlashOpenRouter harness, z-ai/glm-5.3-flash-20260826, Parasail, public 90-day aggregate· OpenRouter benchmark exports83.1%
942GLM 5.3OpenRouter harness, z-ai/glm-5.3-20260816, Cloudflare, public 90-day aggregate· OpenRouter benchmark exports83%
943Qwen3.6 35B A3BOpenRouter harness, qwen/qwen3.6-35b-a3b-20260415, Venice, public 90-day aggregate· OpenRouter benchmark exports83%
944Muse Glimmer 30BOpenRouter harness, meta/muse-glimmer-30b-20260810, Phala, public 90-day aggregate· OpenRouter benchmark exports83%
945Qwen3.5 35B A3BOpenRouter harness, qwen/qwen3.5-35b-a3b-20260224, SiliconFlow, public 90-day aggregate· OpenRouter benchmark exports82.9%
946Qwen3.6 35B A3BOpenRouter harness, qwen/qwen3.6-35b-a3b-20260415, CoreWeave, public 90-day aggregate· OpenRouter benchmark exports82.9%
947GLM 5.3OpenRouter harness, z-ai/glm-5.3-20260816, Morph, public 90-day aggregate· OpenRouter benchmark exports82.9%
948GLM 5.3OpenRouter harness, z-ai/glm-5.3-20260816, CoreWeave, public 90-day aggregate· OpenRouter benchmark exports82.9%
949Gemma 4 31B ITOpenRouter harness, google/gemma-4-31b-it-20260402, CoreWeave, public 90-day aggregate· OpenRouter benchmark exports82.9%
950Qwen3.5 35B A3BOpenRouter harness, qwen/qwen3.5-35b-a3b-20260224, Alibaba, public 90-day aggregate· OpenRouter benchmark exports82.9%

Showing 901–950 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)