Claude Fable 5.2 is the best AI for document analysis. GPT-6.1 Terra scores 51.3 for 5% of the cost, and Kimi K3.1 Pro answers fastest.
Claude Fable 5.2 is the best AI for document analysis. GPT-6.1 Terra scores 51.3 for 5% of the cost, and Kimi K3.1 Pro answers fastest.
32 models · Updated November 2026 · Data from Artificial Analysis
Claude Fable 5.2
Anthropic · Max effort
The highest score of any model that accepts images at this setting.
GPT-6.1 Terra
OpenAI · Max effort
79% of Claude Fable 5.2's score for 5% of the cost.
Kimi K3.1 Pro
Moonshot AI · Standard effort
The quickest full answer from a model scoring 46 or more.
Over 45 s at every tested setting
Claude Fable 5.2
Anthropic · Max effort
The highest score of any model that accepts images at this setting.
GPT-6.1 Terra
OpenAI · Max effort
79% of Claude Fable 5.2's score for 5% of the cost.
Kimi K3.1 Pro
Moonshot AI · Standard effort
The quickest full answer from a model scoring 46 or more.
Each job favors a different model. Pick the one closest to yours.
Contracts and reports that need careful reading
Use
Claude Fable 5.2
Anthropic · Max effort
The strongest reasoning of the models we compared, for documents where a missed clause matters. $345 per 1,000 documents.
Contracts and reports that need careful reading
Use Claude Fable 5.2
The strongest reasoning of the models we compared, for documents where a missed clause matters. $345 per 1,000 documents.
Invoice and form extraction at volume
Use GPT-6.1 Terra
$16.76 per 1,000 documents, about 5% of what Claude Fable 5.2 costs.
Questions about a document while someone waits
Use Kimi K3.1 Pro
A full answer in about 85 seconds for $53.20 per 1,000 documents. With every model held to answers under 45 seconds, GPT-6 Astra is quickest, at about 22 seconds for a score of 51.6.
Staying with OpenAI
Use GPT-6 Astra
The highest-scoring OpenAI model: 56.9 at max effort, with a full answer in 5 minutes. At medium effort it answers in 22 seconds and scores 51.6.
Scanned pages and photos of paper
If the text isn't selectable, it has to be read from the image first. The OCR page covers that step.
Charts and figures inside reports
Explaining a chart needs image reasoning more than document handling.
Each dot is a model. Higher scores better, further left is cheaper, so the best deals sit top left. The cost axis is logarithmic: every step left is a big saving. Hover a dot for its numbers.
Sort by the pillar you care about, and set your monthly volume to see the bill. Costs include each model's thinking tokens at the setting shown.
Claude Fable 5.2. At max effort it scores 64.6 against 59.1 for Claude Opus 5.5, and costs $345 per 1,000 documents against $136. If you need answers in under 45 seconds, Claude Fable 5.2 at high effort scores 61.8 for $150.
Claude Opus 5.5 is the value option: 91% of the score for 40% of the cost. Claude Sonnet 5.5 costs $124 per 1,000 documents and scores 54.5. Claude Haiku 5.5 is the cheap Claude, at $5.20 per 1,000 documents for a score of 36.5. At high effort it answers in 32 seconds and scores 29.9.
All three picks accept at least 1 million tokens, so a long report fits in one go. The limit you'll hit first is cost.
Every page you send is billed as input, and every follow-up question sends the whole document again. For long files, pull out the sections you need first, or ask all your questions in one request.
For routine summaries, GPT-6.1 Sol at medium effort is enough: it answers in about 18 seconds and costs $16.02 per 1,000 documents. Use Claude Fable 5.2 when the summary has to catch nuance, like obligations in a contract or caveats in a research paper. Most of the cost is reading the document, so a shorter summary saves less than you'd expect.
Every model on this page accepts both text and images, so it can read scanned pages as well as digital PDFs. We rank each model at max effort, the setting that gives its best score and the one Artificial Analysis and the labs quote. Cost and answer time are shown at that same setting, so you can see what the top score costs. Switch to Answers under 45 s above to see each model at its strongest setting that gives a full answer within 45 seconds. Effort labels mean different things at different labs, so that view picks each setting by wait time, not by name.
Intelligence is the Artificial Analysis Intelligence Index at that setting. It measures general reasoning, which is what careful reading and extraction depend on.
Cost starts from a short document: 6,000 input tokens for the pages and your instructions, and 4,000 output tokens for the answer. Longer documents cost more in proportion. We scale the output by how much each model wrote, thinking included, in Artificial Analysis's tests, so a model that thinks a lot pays for it.
Speed is the median time to a full answer in the same tests.
Best value is the most intelligence per dollar and fastest is the quickest full answer, both among models scoring 46 or more (within 30% of the leader). Neither pick goes to a model its maker has replaced.
Claude Fable 5.2 at max effort. It scores 64.6 on the Artificial Analysis Intelligence Index, the highest of the models we compared, and costs about $345 per 1,000 documents. For bulk extraction, GPT-6.1 Terra costs $16.76. If you need answers in under 45 seconds, it scores 61.8 at high effort for $150 per 1,000 documents.
Claude Fable 5.2. At max effort it scores 64.6 against 59.1 for Claude Opus 5.5, and costs $345 per 1,000 documents against $136. If you need answers in under 45 seconds, Claude Fable 5.2 at high effort scores 61.8 for $150. Claude Opus 5.5 is the value option: 91% of the score for 40% of the cost. Claude Sonnet 5.5 costs $124 per 1,000 documents and scores 54.5. Claude Haiku 5.5 is the cheap Claude, at $5.20 per 1,000 documents for a score of 36.5. At high effort it answers in 32 seconds and scores 29.9.
Muse Ember, at about $2.12 per 1,000 documents. It scores 36, so check a sample of its extractions before you rely on it.
Kimi K3.1 Pro, with a full answer in about 85 seconds. It scores 46.5, so it suits quick lookups more than careful review. With every model held to answers under 45 seconds, GPT-6 Astra is quickest, at about 22 seconds for a score of 51.6.
For most summaries, GPT-6.1 Sol at medium effort: a full answer in about 18 seconds for $16.02 per 1,000 documents. For summaries that must catch nuance, Claude Fable 5.2 scores highest at 64.6.
Yes. Every model on this page accepts images, so it can read scanned pages directly. For large batches of scans, the OCR page covers the cheapest ways to extract the text first.
Answer three questions and get a pick for each part of your workload, with the monthly cost.