Skip to content
Weights Not Walls
Menu

GDP.pdf

Reported

No description yet: we only describe an eval set when we can do it correctly.

Kind
Published by a third party's leaderboard, collected by Epoch AI. The leaderboard is named and linked.
Link
https://surgehq.ai/benchmarks/gdp-pdf
Score shown
Epoch's column "GDP.pdf score", as a percentage, higher is better. The source gives no error bars.
Results
19 in all: 6 for open-weight models, 13 for closed ones. 5 linked to a model page here.

Best open-weight result: moonshotai/Kimi-K3.1-Pro, 46.6%, number 12 of 19 results.

GDP.pdf results
#ModelWeightsSettingScoreDateSource
1gemini-4-carbon_highGoogle DeepMindclosedeffort high56.4%not datedsurgehq.ai
2muse-wonder-1.0_maxMeta AIclosedeffort max54.8%not datedsurgehq.ai
3gemini-4-argon_highGoogle DeepMindclosedeffort high53.6%not datedsurgehq.ai
4muse-spark-1.3_maxMeta AIclosedeffort max53.0%not datedsurgehq.ai
5gpt-6.1-sol_maxOpenAIclosedeffort max51.6%not datedsurgehq.ai
6claude-sonnet-5-5_maxAnthropicclosedeffort max51.2%not datedsurgehq.ai
7gpt-6-astra_maxOpenAIclosedeffort max50.4%not datedsurgehq.ai
8claude-fable-5-1_maxAnthropicclosedeffort max49.4%not datedsurgehq.ai
9muse-ember-1.0_maxMeta AIclosedeffort max49.2%not datedsurgehq.ai
10claude-opus-5-5_maxAnthropicclosedeffort max48.8%not datedsurgehq.ai
11grok-4.7_xhighxAIclosedeffort xhigh48.8%not datedsurgehq.ai
12moonshotai/Kimi-K3.1-ProEpoch: kimi-k3.1-pro_max, Moonshotopeneffort max46.6%not datedsurgehq.ai
13gemini-3.8-flash_highGoogle DeepMindclosedeffort high44.2%not datedsurgehq.ai
14thinkingmachines/Inkling-2-SmallEpoch: inkling-2-small_xhigh, Thinking Machinesopeneffort xhigh43.6%not datedsurgehq.ai
15qwen3.8-max_xhighAlibabaclosedeffort xhigh42.8%not datedsurgehq.ai
16moonshotai/Kimi-K3Epoch: kimi-k3_max, Moonshotopeneffort max39.8%not datedsurgehq.ai
17zai-org/GLM-5.3Epoch: glm-5.3_max, Z.ai (Zhipu AI)openeffort max36.6%not datedsurgehq.ai
18tencent/Hy4-previewEpoch: hy4-preview_unknown, Tencentopeneffort unknown35.0%not datedsurgehq.ai
19mistral-large-4-0_highMistral AIno page matchopeneffort high32.8%not datedsurgehq.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