Skip to content
Weights Not Walls
Menu

GDPval

Reported

Work tasks from many occupations, where the model's output is compared with an expert's by expert graders.

Kind
Published by a third party's leaderboard, collected by Epoch AI. The leaderboard is named and linked.
Score shown
Epoch's column "Win Rate (%)", as a percentage, higher is better; random guessing scores about 0.0%. The source gives no error bars.
Results
11 in all: 0 for open-weight models, 11 for closed ones. 0 linked to a model page here.

No open-weight model has a result on this set.

GDPval results
#ModelWeightsSettingScoreDateSource
1gpt-5.2-2025-12-11_noneOpenAIclosedeffort none49.7%not datedvia Epoch AI
2claude-opus-4-5-20251101Anthropiccloseddefault45.5%not datedvia Epoch AI
3claude-opus-4-1-20250805Anthropiccloseddefault43.6%not datedvia Epoch AI
4claude-sonnet-4-5-20250929Anthropiccloseddefault42.5%not datedvia Epoch AI
5gemini-3-pro-previewGoogle DeepMindcloseddefault40.3%not datedvia Epoch AI
6gpt-5-2025-08-07_mediumOpenAIclosedeffort medium34.8%not datedvia Epoch AI
7o3-2025-04-16_mediumOpenAIclosedeffort medium30.8%not datedvia Epoch AI
8o4-mini-2025-04-16_highOpenAIclosedeffort high25.3%not datedvia Epoch AI
9gemini-2.5-proGoogle DeepMindcloseddefault23.3%not datedvia Epoch AI
10grok-4-0709_highxAIclosedeffort high21.1%not datedvia Epoch AI
11gpt-4o-2024-11-20OpenAIcloseddefault9.9%not datedvia Epoch AI

Open-weight rows are shaded. Dates are when the result was run or evaluated, where the data says. Data from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Epoch AI. Method