Vision & documents

GDP.pdf leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on GDP.pdf is 32.2% by GPT-6 Astra (xhigh). 3 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: artificial-analysis.

#	Model	Score	Source
1	GPT-6 Astra (xhigh)	32.2%	measured
2	GPT-6.1 Sol (high)	32.0%	measured
3	GPT-6.1 Sol (xhigh)	31.8%	measured
4	GPT-6 Astra (high)	31.0%	measured
5	GPT-6 Astra (max)	31.0%	measured
6	GPT-6.1 Sol (max)	31.0%	measured
7	GPT-6 Astra (medium)	30.4%	measured
8	Claude Opus 5.5 (xhigh with fallback)	29.1%	estimated ± 4.0 pp, high confidence
9	Claude Opus 5.5 (high with fallback)	27.6%	estimated ± 4.0 pp, high confidence
10	Claude Opus 5.5 (medium with fallback)	27.6%	estimated ± 4.0 pp, high confidence
11	Muse Spark 1.3 (max)	26.6%	measured
12	Claude Fable 5.1 (max with fallback)	26.2%	measured
13	Claude Opus 5.5 (max with fallback)	26.2%	measured
14	Claude Sonnet 5.5 (max with fallback)	25.8%	measured
15	GPT-6 Sol (max)	25.2%	measured
16	Qwen3.8 Max (0902)	22.8%	measured
17	GPT-6 Luna (max)	22.8%	measured
18	Kimi K3 (max)	22.0%	measured
19	Gemini 4 Argon (high)	21.8%	measured
20	Gemini 3.8 Flash (high)	21.0%	measured
21	Grok 4.7 (xhigh)	20.0%	measured
22	MiMo-V2.6-Pro	19.2%	measured
23	Mistral Large 4 Preview	18.6%	measured
24	Qwen3.8 27B (xhigh)	16.6%	measured
25	GLM-5.3-Flash	15.4%	measured
26	Step 5 Preview	14.8%	measured
27	Inkling (xhigh)	12.8%	measured
28	DeepSeek V4.1 Flash (max)	12.8%	measured
29	GLM-5.3 (max)	11.2%	measured
30	Muse Glimmer (high)	10.0%	measured
31	MiniMax-M3	9.8%	measured
32	K2 Horizon 375B A23B	7.4%	measured
33	Nemotron 3 Ultra	5.0%	measured

GDP.pdf leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Vision & documents