Tests explained · Test of real work

Pulling information out of documents into tables

Reading documents and filling in a table with the right figures, including cases that need reasoning.

Run by

DTBench authors

How to read it

The share of tasks the model got right. The thin line, where shown, is the likely range.

Used for

5 tasks on this site.

How the models did

Share of tasks done correctly · Higher is better

Claude Fable 5.1 leads, followed by GPT-6 Astra and Gemini 3.1 Pro.

Claude Fable 5.1 · Anthropic98%

GPT-6 Astra · OpenAI97%

Gemini 3.1 Pro · Google DeepMind97%

GPT-5.6 Sol · OpenAI96%

Gemini 3.8 Flash · Google DeepMind96%

Gemini 3.6 Flash · Google DeepMind95%

DeepSeek V4 Pro · DeepSeek94%

Claude Sonnet 5 · Anthropic92%

Kimi K3 · Moonshot AI91%

DeepSeek V4.1 Flash · DeepSeek90%

GPT-5.6 Luna · OpenAI89%

Gemini 3.5 Flash-Lite · Google DeepMind83%

Claude Haiku 4.5 · Anthropic74%

Claude Opus 5.5 · AnthropicNot tested yet

GPT-6 Luna · OpenAINot tested yet

GPT-6 Sol · OpenAINot tested yet

Source: DTBench authors via Epoch AI · CC BY 4.0

Where we use it

Each task page labels this test by how closely it matches the task.

Task	How close
Best AI for documents, PDFs and notes	Tests this task
Best AI for reading scanned documents	Tests a related skill
Best AI for financial analysis	Tests a related skill
Best AI for Excel and spreadsheets	Tests a related skill
Best AI for data analysis	Tests a related skill