GPQA Diamond leaderboard
As of November 8, 2026, GPT-6 Astra (OpenAI) leads GPQA Diamond at 96%, ahead of GPT-5.6 Sol at 94.6% and Gemini 3.1 Pro Preview at 94.3%. 11 systems ranked. Every score here is published by the model's own lab; each row links to the page it comes from.
What is GPQA Diamond? GPQA Diamond is a set of 198 graduate-level multiple-choice questions in biology, physics and chemistry, written so that skilled non-experts with web access still get them wrong. Made by David Rein, Samuel R. Bowman and co-authors (New York University), November 2023.
Top models on GPQA Diamond, November 2026
| # | Model | Score | Price / 1M | Context | Source |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra OpenAI | as reported | $10 / $50 | 1.05M | Vendor-reported OpenAI |
| 2 | GPT-5.6 Sol OpenAI | as reported | $4 / $20 | 1.05M | Vendor-reported OpenAI |
| 3 | Gemini 3.1 Pro Preview Preview Google · listed as “Gemini 3.1 Pro” | no tools, thinking high | $2 / $12 | 1M | Vendor-reported Google DeepMind |
| 4 | Kimi K3.1 Pro Moonshot (Kimi) | max effort | $2 / $10 | 1M | Vendor-reported Moonshot AI |
| 5 | Kimi K3 Moonshot (Kimi) | max effort | $3 / $15 | 1M | Vendor-reported Moonshot AI |
| 6 | Qwen3.8 Max Alibaba (Qwen) | as reported | $2 / $6 | 1M | Vendor-reported Alibaba (Qwen) |
| 7 | DeepSeek V4 Pro DeepSeek | pass@1 | $1.32 / $3.96 | 1M | Vendor-reported DeepSeek |
| 7 | Qwen3.7 Max Alibaba (Qwen) | as reported | $2.5 / $7.5 | 1M | Vendor-reported Alibaba (Qwen) |
| 9 | Gemini 3 Pro Preview Retired Google · listed as “Gemini 3 Pro” | no tools, thinking high | $2 / $12 | 1M | Vendor-reported Google DeepMind |
| 10 | DeepSeek V4.1 Flash DeepSeek | pass@1 | $0.3 / $1.2 | 1M | Vendor-reported DeepSeek |
| 11 | DeepSeek V4 Flash Deprecated DeepSeek | pass@1 | $0.44 / $1.32 | 1M | Vendor-reported DeepSeek |
GPQA Diamond: score against price
Can you trust GPQA Diamond scores?
No independent leaderboard publishes GPQA Diamond results for these models yet, so every number on this page is the lab's own report of its own model. Labs choose the harness, the effort level and the subset they report, and they rarely publish failures. Read the setting under each score before comparing two rows.
- With 198 questions, one question is half a point: differences of a point or two between models are noise.
- Even PhD experts in the field scored about 65% when the set was built, and some questions have known flaws, so the ceiling is below 100%.
- There is no official leaderboard, and labs choose their own prompting and sampling.
How GPQA Diamond works
The task
A four-option multiple-choice question written by a domain expert; the model has to pick the correct option.
The score
Plain accuracy; random guessing scores 25%. Diamond keeps only questions where both expert validators agreed with the answer and most non-experts failed.
Test set
198 questions, as published by the authors.
Made by
David Rein, Samuel R. Bowman and co-authors (New York University), first published November 2023.
Official sources
- github.com/idavidrein/gpqa
- Paper: arxiv.org/abs/2311.12022
- Definition read on 2026-11-08
Sources and freshness
11 published scores from 6 sources. Each was read on its source page; the oldest reading on this page is from November 8, 2026. Prices beside each score come from our model price index and move when a vendor reprices.
- Vendor-reported DeepSeek
- Vendor-reported Alibaba (Qwen)
- Vendor-reported Google DeepMind
- Vendor-reported OpenAI
- Vendor-reported Moonshot AI
- Vendor-reported Moonshot AI
GPQA Diamond FAQ
Answers derived from the 11 published scores on this page and the current price index.