Last verified 2026-11-08

Vision and documentsVendor-reportedPrices verified

GDP.pdf leaderboard

As of November 8, 2026, Gemini 3.8 Flash (Google DeepMind) leads GDP.pdf at 35%, ahead of Kimi K3.1 Pro at 34.6% and Gemini 3.7 Flash at 34%. 7 systems ranked. Every score here is published by the model's own lab; each row links to the page it comes from.

See the ranking →All benchmarks →Model prices →

Top score

35%

Gemini 3.8 Flash

Systems ranked

across 2 settings

Independent rows

0/8

all vendor-reported

Cheapest near top

$0.481/1M

Gemini 3.8 Flash · within 5 pts

With API prices

7/7

rates from our price index

Newest score

Nov 2

2026 · newest result

What is GDP.pdf? GDP.pdf tests whether a model can answer the questions working professionals ask about real PDF documents: leases, datasheets, clinical guidelines, construction plans. Made by Surge AI, July 2026.

§ 01 / LEADERBOARD

Top models on GDP.pdf, November 2026

7 entries · Default setting

Tools are off by default in the official harness; rows here either say no tools or do not state tool use.

#	Model	Score	$ / task	Price / 1M	Context	Source
1	Gemini 3.8 Flash Google DeepMind	35% all-pass rate	—	$0.75 / $3.75	1M	Vendor-reported Google DeepMind 2026-09-02
2	Kimi K3.1 Pro Moonshot (Kimi)	34.6% max effort, no tools	—	$2 / $10	1M	Vendor-reported Moonshot AI 2026-11-02
3	Gemini 3.7 Flash Deprecated Google DeepMind	34% all-pass rate	—	$0.75 / $3.75	1M	Vendor-reported Google DeepMind 2026-09-02
4	GPT-6 Astra OpenAI	32.2% xhigh effort	$1.91	$10 / $50	1.05M	Vendor-reported OpenAI 2026-09-29
5	GPT-6.1 Sol OpenAI	32% high effort	$0.35	$2 / $10	1.05M	Vendor-reported OpenAI 2026-09-29
6	GPT-6 Sol OpenAI	28% high effort	$0.35	$2 / $10	1.05M	Vendor-reported OpenAI 2026-09-29
7	Gemini 3.6 Flash Google	22% as reported	—	$0.75 / $3.75	1M	Vendor-reported Google DeepMind 2026-08-13

§ 02 / OTHER SETTINGS

GDP.pdf under other settings

Labs report GDP.pdf under different settings, and a score from one setting is not comparable with a score from another. Each setting has its own table, ranked on its own.

02.1

With tools

1 entry

The model may run Python on the document.

#	Model	Score	Price / 1M	Context	Source
1	Kimi K3.1 Pro Moonshot (Kimi)	85.7% max effort, with Python	$2 / $10	1M	Vendor-reported Moonshot AI 2026-11-02

§ 03 / SCORE VS PRICE

GDP.pdf: score against price

The dashed line joins the models that no cheaper model beats: Gemini 3.8 Flash (35%, $0.481 per 1M tokens). Anything below the line costs more for a lower score. Price is the effective rate per 1M tokens on current list prices: 92% input, 8% output, 82% of input served from cache.

§ 04 / REASONING EFFORT

What more reasoning buys

Some publishers ran the same model at several reasoning-effort levels and printed the score and the cost per task for each. Higher effort spends more tokens per task; this is what it bought on GDP.pdf.

Model	Score by effort level ($ per task)	Source
GPT-6 Astra	low 30.4% $1.70 medium 30.4% $1.72 high 31% $1.79 xhigh 32.2% $1.91 max 31% $2.08	OpenAI
GPT-6.1 Sol	low 27% $0.33 medium 30% $0.34 high 32% $0.35 xhigh 31.8% $0.37 max 31% $0.42	OpenAI
GPT-6 Sol	low 21.8% $0.33 medium 25.4% $0.34 high 28% $0.35 xhigh 23.8% $0.37 max 24.8% $0.43	OpenAI

§ 05 / PROVENANCE

Can you trust GDP.pdf scores?

No independent leaderboard publishes GDP.pdf 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.

!Items were chosen because frontier models failed them, which makes the test hard by construction: in the paper the best model passed 30.7% of items.
!Grading uses a model judge on the rubrics; labs that use another judge or report the graded score instead of the strict pass rate are not comparable.
!GDP.pdf is not OpenAI's GDPval despite the similar name.

§ 06 / HOW IT WORKS

How GDP.pdf works

The task

Each item pairs a professional PDF with a realistic expert question. Questions were kept only when at least two frontier models failed them in a way that mattered.

The score

Each item has a rubric of yes-or-no criteria. The headline is the strict pass rate: an item passes only when every criterion passes; a graded rubric score is reported beside it.

Test set

100 items across ten professional fields, as published by the authors.

Made by

Surge AI, first published July 2026.

Official sources

§ 07 / SOURCES

Sources and freshness

8 published scores from 4 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 OpenAI 3 scores · newest 2026-09-29
Vendor-reported Google DeepMind 2 scores · newest 2026-09-02
Vendor-reported Moonshot AI 2 scores · newest 2026-11-02
Vendor-reported Google DeepMind 1 score · newest 2026-08-13

GDP.pdf FAQ

Answers derived from the 8 published scores on this page and the current price index.

Q · 01

Which AI model has the highest GDP.pdf score?

Gemini 3.8 Flash with 35% (all-pass rate), as reported by Google DeepMind on September 2, 2026. Next: Kimi K3.1 Pro at 34.6% and Gemini 3.7 Flash at 34%.

Q · 02

What is the cheapest model near the top of GDP.pdf?

The leader itself. No priced model within 5 points of Gemini 3.8 Flash is cheaper: it runs at $0.481 per 1M tokens on our effective blend ($0.75 input, $3.75 output).

Q · 03

Are these GDP.pdf scores independently verified?

Not yet. Every score here is the model maker's own report. Each row names the lab and links to the page where it published the number.

Q · 04

What does GDP.pdf measure?

GDP.pdf tests whether a model can answer the questions working professionals ask about real PDF documents: leases, datasheets, clinical guidelines, construction plans. Each item pairs a professional PDF with a realistic expert question. Questions were kept only when at least two frontier models failed them in a way that mattered.

Q · 05

How is the GDP.pdf score calculated?

Each item has a rubric of yes-or-no criteria. The headline is the strict pass rate: an item passes only when every criterion passes; a graded rubric score is reported beside it. The test set has 100 items across ten professional fields.

Q · 06

How current is this GDP.pdf leaderboard?

The newest score was published on November 2, 2026. Every row shows the date its source published it, and we last read all of them on November 8, 2026. Prices come from our model price index, where each rate carries the date it was read on the vendor's page.