Reported by GDP.pdf

GDP.pdf

Name: GDP.pdf results
Creator: GDP.pdf
License: https://surgehq.ai/benchmarks/gdp-pdf

Rubric score on 100 real professional tasks that need reasoning over PDFs (finance, legal, insurance, engineering, HR and more).

Last updated 1 Oct 2026

Results dated: 6 Jun 2026 to 1 Oct 2026
Results: 49 configurations of 32 models
Unit: % of rubric
Licence: Published with permission (Surge AI results via Epoch AI)

Professional document tasks: GPT-6 Luna

Top 15 of 32 results · % of rubric, higher is better. Choose a model to highlight it.Clear highlight

1 GPT-5.6 Sol (max reasoning)OpenAI 30.7%
2 Claude Opus 5.5 (max reasoning)Anthropic 30.6%
3 Claude Fable 5Anthropic 30.0%
4 Claude Fable 5.1 (high reasoning)Anthropic 29.6%
5 Muse Spark 1.3 (extra-high reasoning)Meta 27.6%
6 GPT-6 Sol (max reasoning)OpenAI 26.4%
7 GPT-5.5 (extra-high reasoning)OpenAI 26.0%
8 GPT-5.6 Terra (medium reasoning)OpenAI 24.7%
9 Claude Opus 4.8 (max reasoning)Anthropic 24.0%
9 Claude Opus 5 (max reasoning)Anthropic 24.0%
11 Gemini 3.7 Flash (high reasoning)Google 23.8%
12 Gemini 3.8 Flash (medium reasoning)Google 23.4%
13 Qwen3.8-Max (0902, extra-high reasoning)Alibaba 23.2%
14 GPT-6 Luna (max reasoning)OpenAI 23.0%
15 GPT-5.6 Luna (medium reasoning)OpenAI 22.7%

Full results

GDP.pdf: professional document tasks, % of rubric, higher is better
#	Model	Professional document tasks % of rubric, higher is better	Price $ per million tokens, in / out
1	GPT-5.6 Sol (max reasoning)OpenAI · best of 2 settings	30.7%	$4 / $20
2	Claude Opus 5.5 (max reasoning)Anthropic	30.6%	$4 / $20
3	Claude Fable 5Anthropic · best of 2 settings	30.0%	$10 / $50
4	Claude Fable 5.1 (high reasoning)Anthropic · best of 2 settings	29.6%	$10 / $50
5	Muse Spark 1.3 (extra-high reasoning)Meta · best of 2 settings	27.6%	$1.25 / $4.25
6	GPT-6 Sol (max reasoning)OpenAI	26.4%	$2 / $10
7	GPT-5.5 (extra-high reasoning)OpenAI	26.0%	$5 / $30
8	GPT-5.6 Terra (medium reasoning)OpenAI · best of 2 settings	24.7%	$2 / $12
9	Claude Opus 4.8 (max reasoning)Anthropic	24.0%	$5 / $25
9	Claude Opus 5 (max reasoning)Anthropic	24.0%	$5 / $25
11	Gemini 3.7 Flash (high reasoning)Google · best of 2 settings	23.8%	$1.50 / $7.50
12	Gemini 3.8 Flash (medium reasoning)Google · best of 2 settings	23.4%	$1.50 / $7.50
13	Qwen3.8-Max (0902, extra-high reasoning)Alibaba	23.2%	$2 / $6
14	GPT-6 Luna (max reasoning)OpenAI	23.0%	$0.10 / $0.50
15	GPT-5.6 Luna (medium reasoning)OpenAI · best of 2 settings	22.7%	$0.20 / $1.20
16	Claude Opus 4.7 (max reasoning)Anthropic	21.0%	$5 / $25
17	Kimi K3 (max reasoning)Moonshot AI	19.0%	$3 / $15
18	Claude Sonnet 4.6 (max reasoning)Anthropic	18.0%	$3 / $15
19	Grok 4.6 (extra-high reasoning)xAI · best of 2 settings	17.2%	$2 / $6
20	Gemini 3.1 Pro Preview (high reasoning)Google · best of 2 settings	17.0%	$2 / $12
21	Qwen3.8-Flash (extra-high reasoning)Alibaba	16.6%	$0.15 / $0.47
22	Muse Spark 1.2 (extra-high reasoning)Meta · best of 2 settings	16.0%	$1.25 / $4.25
23	Muse Spark 1.1 (medium reasoning)Meta · best of 2 settings	15.0%	$1.25 / $4.25
24	Gemini 3.5 Flash (medium reasoning)Google · best of 2 settings	14.0%	$1.50 / $9
24	Gemini 3.6 Flash (high reasoning)Google · best of 2 settings	14.0%	$1.50 / $7.50
24	Grok 4.5 (high reasoning)xAI	14.0%	$2 / $6
24	GLM-5.3-Flash (max reasoning)Z.ai	14.0%	$0.15 / $0.50
28	Kimi K2.6Moonshot AI	12.0%	$0.95 / $4
29	Gemini 3.5 Flash-Lite (high reasoning)Google	10.0%	$0.30 / $2.50
29	Gemini 3 Flash Preview (high reasoning)Google · best of 2 settings	10.0%	$0.50 / $3
31	Grok 4.3 (high reasoning)xAI · best of 2 settings	8.0%	$1.25 / $2.50
32	Nova 2 Pro Preview (no reasoning)Amazon · best of 2 settings	2.0%	–

Swipe the table sideways for more columns.

Ranks follow the score as shown, so equal numbers share a rank. Each model is shown at its best setting; show every setting. Results as published by GDP.pdf; we do not re-run them.

Progress over time

Professional document tasks, plotted at each model's release date, higher is better. Each model counts once, at its best setting. Scores are today's, not launch day's; 1 model without a known release date is left off. Hover or tap a mark for its model.

Best score so far

Every model at its release date. The line moves each time a new release beats every earlier one, and each new best is named where there is room.

Best score by lab

Each lab's best so far, stepping whenever one of its releases beats its own earlier models. The 6 labs with the best scores today; 2 others not drawn.

OpenAIAnthropicMetaGoogleAlibabaMoonshot AI

What it measures

Rubric score on 100 real professional tasks that need reasoning over PDFs (finance, legal, insurance, engineering, HR and more).

What it does not measure

Not your documents; 100 tasks, so small differences are noise.

Source

GDP.pdf · Methodology

Surge AI; collected by Epoch AI. Licence: Published with permission (Surge AI results via Epoch AI).