GPQA Diamond
Graduate-level science questions, 198-question Google-proof subset. Measured by Vals AI: Vals AI standardized run. This benchmark is archived as saturated and is no longer run on new releases.
48models scored
measured 2026-09-01
measured 2026-09-01
Why this is archived
Vals archived this benchmark as saturated. The page describes both zero-shot and five-shot chain-of-thought approaches but does not state which produced the published scores, so treat small gaps as noise.
Coverage
GPQA Diamond ranks 138 models; this table covers the 48 of them whose pricing we have verified against the provider's own page, so 90 scored models are not shown.
Source
Vals AI — GPQA Diamond, read 2026-11-20
Standings
| Model | Maker | Score | Blended $/MTok | Score per $ |
|---|---|---|---|---|
| 1Gemini 3.1 Pro Preview Preview 02/26 Preview | 95.45% | $4.50 | 21.2 | |
| 2gpt-5.6-sol | OpenAI | 95.20% | $8.00 | 11.9 |
| 3Grok 4.6 | xAI | 94.70% | $3.00 | 31.6 |
| 4Gemini 3.8 Flash | 94.44% | $1.50 | 63.0 | |
| 5Claude Fable 5.1 | Anthropic | 93.43% | $20.00 | 4.7 |
| 6Claude Opus 5 | Anthropic | 93.43% | $10.00 | 9.3 |
| 7Gemini 3.6 Flash | 93.43% | $1.50 | 62.3 | |
| 8Claude Fable 5 | Anthropic | 93.18% | $20.00 | 4.7 |
| 9gpt-5.5 | OpenAI | 93.18% | $11.25 | 8.3 |
| 10Grok 4.5 | xAI | 92.93% | $3.00 | 31.0 |
| 11Kimi K3 | Moonshot | 92.93% | $6.00 | 15.5 |
| 12Claude Opus 4.8 | Anthropic | 92.42% | $10.00 | 9.2 |
| 13DeepSeek V4 Pro 0813 | DeepSeek | 92.42% | $1.98 | 46.7 |
| 14gpt-5.6-luna | OpenAI | 91.67% | $0.45 | 203.7 |
| 15gpt-5.4 xhigh | OpenAI | 91.67% | $5.63 | 16.3 |
| 16gpt-5.2 | OpenAI | 91.67% | $4.81 | 19.0 |
| 17Grok 4.3 | xAI | 91.41% | $1.56 | 58.5 |
| 18gpt-5.6-terra | OpenAI | 90.91% | $4.50 | 20.2 |
| 19Claude Opus 4.7 | Anthropic | 90.15% | $10.00 | 9.0 |
| 20Claude Opus 4.6 Thinking | Anthropic | 89.65% | $10.00 | 9.0 |
| 21Kimi K2.6 | Moonshot | 89.14% | $1.71 | 52.1 |
| 22Claude Sonnet 5 | Anthropic | 88.89% | $4.00 | 22.2 |
| 23GLM-5.3 | Z.ai | 88.13% | $2.15 | 41.0 |
| 24Gemini 3 Flash Preview 12/25 Preview | 87.88% | $1.13 | 78.1 | |
| 25gpt-5.1 | OpenAI | 86.62% | $3.44 | 25.2 |
| 26GLM-5.3-Flash | Z.ai | 86.36% | $0.237 | 363.6 |
| 27Claude Opus 4.5 Thinking | Anthropic | 85.86% | $10.00 | 8.6 |
| 28Claude Sonnet 4.6 | Anthropic | 85.61% | $6.00 | 14.3 |
| 29gpt-5 | OpenAI | 85.61% | $3.44 | 24.9 |
| 30GLM-5.2 | Z.ai | 85.61% | $2.15 | 39.8 |
| 31GLM-5.1 | Z.ai | 84.52% | $2.15 | 39.3 |
| 32o3 | OpenAI | 84.09% | $3.50 | 24.0 |
| 33Gemini 3.5 Flash-Lite | 83.84% | $0.85 | 98.6 | |
| 34GLM-5 | Z.ai | 83.33% | $1.55 | 53.8 |
| 35gpt-5.4-mini | OpenAI | 83.08% | $1.69 | 49.2 |
| 36Claude Sonnet 4.5 Thinking | Anthropic | 81.63% | $6.00 | 13.6 |
| 37Gemini 3.1 Flash-Lite Preview | 81.06% | $0.563 | 144.1 | |
| 38gpt-5-mini | OpenAI | 80.30% | $0.688 | 116.8 |
| 39GLM-4.7 | Z.ai | 80.05% | $1.00 | 80.0 |
| 40gpt-5.4-nano | OpenAI | 77.53% | $0.463 | 167.6 |
| 41Claude Opus 4.1 Thinking Retired | Anthropic | 76.26% | $30.00 | 2.5 |
| 42Claude Sonnet 4 Thinking Retired | Anthropic | 75.00% | $6.00 | 12.5 |
| 43o4-mini Retired | OpenAI | 74.50% | $1.93 | 38.7 |
| 44GLM-4.6 | Z.ai | 74.50% | $1.00 | 74.5 |
| 45Claude Haiku 4.5 Thinking | Anthropic | 72.22% | $2.00 | 36.1 |
| 46gpt-4.1 | OpenAI | 65.40% | $3.50 | 18.7 |
| 47gpt-5-nano | OpenAI | 63.38% | $0.138 | 460.9 |
| 48gpt-4o 2024-11-20 | OpenAI | 53.79% | $4.38 | 12.3 |