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
601DeepSeek V4 FlashOpenRouter harness, deepseek/deepseek-v4-flash-20260423, Phala, public 90-day aggregate· OpenRouter benchmark exports87.4%
602GPT-6 LunaOpenRouter harness, openai/gpt-6-luna-20260922, auto-routing, public 90-day aggregate· OpenRouter benchmark exports87.4%
603MiniMax M3.1 FlashOpenRouter harness, minimax/minimax-m3.1-flash-20261019, Nebius, public 90-day aggregate· OpenRouter benchmark exports87.4%
604Qwen3.7 PlusOpenRouter harness, qwen/qwen3.7-plus-20260602, Alibaba, public 90-day aggregate· OpenRouter benchmark exports87.4%
605MiMo V2.5 ProOpenRouter harness, xiaomi/mimo-v2.5-pro-20260422, StreamLake, public 90-day aggregate· OpenRouter benchmark exports87.4%
606DeepSeek V4 FlashOpenRouter harness, deepseek/deepseek-v4-flash-20260423, Wafer, public 90-day aggregate· OpenRouter benchmark exports87.3%
607DeepSeek V4.1 FlashOpenRouter harness, deepseek/deepseek-v4.1-flash-20260910, Ionstream, public 90-day aggregate· OpenRouter benchmark exports87.3%
608Qwen3.5 397B A17BOpenRouter harness, qwen/qwen3.5-397b-a17b-20260216, Alibaba, public 90-day aggregate· OpenRouter benchmark exports87.3%
609Claude Sonnet 5OpenRouter harness, anthropic/claude-sonnet-5-20260630, Anthropic, public 90-day aggregate· OpenRouter benchmark exports87.3%
610MiniMax M3.1 FlashOpenRouter harness, minimax/minimax-m3.1-flash-20261019, DigitalOcean, public 90-day aggregate· OpenRouter benchmark exports87.3%
611Qwen3.5 397B A17BOpenRouter harness, qwen/qwen3.5-397b-a17b-20260216, auto-routing, public 90-day aggregate· OpenRouter benchmark exports87.2%
612GLM 5.3OpenRouter harness, z-ai/glm-5.3-20260816, Wafer, public 90-day aggregate· OpenRouter benchmark exports87.2%
613DeepSeek V4.1 FlashOpenRouter harness, deepseek/deepseek-v4.1-flash-20260910, OpenInference, public 90-day aggregate· OpenRouter benchmark exports87.2%
614Kimi K2.5OpenRouter harness, moonshotai/kimi-k2.5-0127, DeepInfra, public 90-day aggregate· OpenRouter benchmark exports87.2%
615Kimi K2.5OpenRouter harness, moonshotai/kimi-k2.5-0127, Moonshot AI, public 90-day aggregate· OpenRouter benchmark exports87.2%
616Inkling SmallOpenRouter harness, thinkingmachines/inkling-small-20260730, BaseTen, public 90-day aggregate· OpenRouter benchmark exports87.2%
617Qwen 3.8 27BOpenRouter harness, qwen/qwen3.8-27b-20260814, Morph, public 90-day aggregate· OpenRouter benchmark exports87.2%
618Kimi K2.7 CodeOpenRouter harness, moonshotai/kimi-k2.7-code-20260612, GMICloud, public 90-day aggregate· OpenRouter benchmark exports87.2%
619MiMo V2.6 ProOpenRouter harness, xiaomi/mimo-v2.6-pro-20260921, GMICloud, public 90-day aggregate· OpenRouter benchmark exports87.2%
620Qwen3.8 MaxOpenRouter harness, qwen/qwen3.8-2.4t-a95b-20260812, DeepInfra, public 90-day aggregate· OpenRouter benchmark exports87.1%
621Qwen3.5 397B A17BOpenRouter harness, qwen/qwen3.5-397b-a17b-20260216, AtlasCloud, public 90-day aggregate· OpenRouter benchmark exports87.1%
622DeepSeek V4 Flash 0731OpenRouter harness, deepseek/deepseek-v4-flash-20260731, auto-routing, public 90-day aggregate· OpenRouter benchmark exports87.1%
623Claude Haiku 5.5OpenRouter harness, anthropic/claude-haiku-5.5-20261007, Amazon Bedrock, public 90-day aggregate· OpenRouter benchmark exports87.1%
624Kimi K2.7 CodeOpenRouter harness, moonshotai/kimi-k2.7-code-20260612, auto-routing, public 90-day aggregate· OpenRouter benchmark exports87.1%
625Claude Haiku 5.5OpenRouter harness, anthropic/claude-haiku-5.5-20261007, Anthropic, public 90-day aggregate· OpenRouter benchmark exports87%
626DeepSeek V4 FlashOpenRouter harness, deepseek/deepseek-v4-flash-20260423, CoreWeave, public 90-day aggregate· OpenRouter benchmark exports87%
627GLM 5.3 FlashOpenRouter harness, z-ai/glm-5.3-flash-20260826, Venice, public 90-day aggregate· OpenRouter benchmark exports87%
628DeepSeek V4 FlashOpenRouter harness, deepseek/deepseek-v4-flash-20260423, Parasail, public 90-day aggregate· OpenRouter benchmark exports87%
629Qwen3.8 MaxOpenRouter harness, qwen/qwen3.8-2.4t-a95b-20260812, Together, public 90-day aggregate· OpenRouter benchmark exports87%
630Claude Opus 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.87%
631Grok 4· Epoch AI Benchmarking Hub data87%
632GLM 5.3OpenRouter harness, z-ai/glm-5.3-20260816, DeepInfra, public 90-day aggregate· OpenRouter benchmark exports87%
633DeepSeek V4 Flash 0731OpenRouter harness, deepseek/deepseek-v4-flash-20260731, Sail Research, public 90-day aggregate· OpenRouter benchmark exports87%
634GLM 5.3OpenRouter harness, z-ai/glm-5.3-20260816, Z.AI, public 90-day aggregate· OpenRouter benchmark exports87%
635Qwen3.5 397B A17BOpenRouter harness, qwen/qwen3.5-397b-a17b-20260216, Venice, public 90-day aggregate· OpenRouter benchmark exports87%
636MiMo V2.6 ProOpenRouter harness, xiaomi/mimo-v2.6-pro-20260921, PrimaLabs, public 90-day aggregate· OpenRouter benchmark exports87%
637Claude Sonnet 5OpenRouter harness, anthropic/claude-sonnet-5-20260630, Google, public 90-day aggregate· OpenRouter benchmark exports87%
638GPT 5.1OpenRouter harness, openai/gpt-5.1-20251113, OpenAI, public 90-day aggregate· OpenRouter benchmark exports87%
639MiniMax M3.1 FlashOpenRouter harness, minimax/minimax-m3.1-flash-20261019, ModelRun, public 90-day aggregate· OpenRouter benchmark exports87%
640Claude Opus 4.564K thinking · interleaved scratchpads · 200K context · avg@5· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.87%
641Qwen3.8 MaxOpenRouter harness, qwen/qwen3.8-2.4t-a95b-20260812, Novita, public 90-day aggregate· OpenRouter benchmark exports86.9%
642DeepSeek V4.1 FlashOpenRouter harness, deepseek/deepseek-v4.1-flash-20260910, Fireworks, public 90-day aggregate· OpenRouter benchmark exports86.9%
643Claude Haiku 5.5OpenRouter harness, anthropic/claude-haiku-5.5-20261007, Claude Platform on AWS, public 90-day aggregate· OpenRouter benchmark exports86.9%
644Gemini 3.1 Flash LiteGemini 3.1 Flash-Lite Model Card · high thinking · default sampling · single attempt/pass@1 · no tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.86.9%
645DeepSeek V4 FlashOpenRouter harness, deepseek/deepseek-v4-flash-20260423, DeepInfra, public 90-day aggregate· OpenRouter benchmark exports86.9%
646Inkling SmallOpenRouter harness, thinkingmachines/inkling-small-20260730, Together, public 90-day aggregate· OpenRouter benchmark exports86.9%
647GPT-5.4 minixhigh· Epoch AI Benchmarking Hub data86.9%
648GPT 5.1OpenRouter harness, openai/gpt-5.1-20251113, Azure, public 90-day aggregate· OpenRouter benchmark exports86.9%
649Claude Opus 4.6OpenRouter harness, anthropic/claude-4.6-opus-20260205, Azure, public 90-day aggregate· OpenRouter benchmark exports86.9%
650DeepSeek V4 ProOpenRouter harness, deepseek/deepseek-v4-pro-20260423, SiliconFlow, public 90-day aggregate· OpenRouter benchmark exports86.9%

Showing 601–650 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)