Best AI model for pulling data out of documents
We never take money for a place in a ranking. Read how we make money.
The verdict
- Scores 67 on support, #12 of 30 models we track for this job.
- $0.65 per 1,000 documents.
- Also the cheapest model that passes our test for this job.
- Scores 97 on support, the highest of 30 models we track for this job.
- $10.00 per 1,000 documents: the extra $9.35 per 1,000 buys 30 more points.
- Reads up to 1,049K tokens at once.
Extraction rewards models that follow a format to the letter. It is a bulk job, so a model that passes our bar at a low price usually wins.
How we tested
Until our rig’s data-extraction task runs, we use the support score (instruction following and multi-turn) as the stand-in. A document is about 3,000 tokens in and 400 out.
Scores are provisional until our own test rig runs these tasks. A model “passes our test” for this job at a score of 50 or more.
Scores and cost at your volume
| Model | Score | Passes | 1k documents a month | 10k documents a month | 100k documents a month |
|---|---|---|---|---|---|
| Gemini 4 Argon | 97 | Yes | $10.00 | $100 | $1000 |
| Claude Opus 5.5 | 90 | Yes | $20.00 | $200 | $2000 |
| Claude Opus 5 | 84 | Yes | $25.00 | $250 | $2500 |
| Claude Fable 5.1 | 83 | Yes | $50.00 | $500 | $5000 |
| Muse Wonder | 79 | Yes | $20.00 | $200 | $2000 |
| Gemini 3.8 Flash | 78 | Yes | $3.75 | $37.50 | $375 |
| Kimi K3 · open | 73 | Yes | $15.00 | $150 | $1500 |
| Qwen3.8 Max | 72 | Yes | $8.40 | $84.00 | $840 |
| Muse Spark 1.3 | 71 | Yes | $5.45 | $54.50 | $545 |
| Claude Sonnet 5.5 | 69 | Yes | $10.00 | $100 | $1000 |
| GLM-5.3 · open | 67 | Yes | $5.96 | $59.60 | $596 |
| GLM-5.3-Flash · open | 67 | Yes | $0.65 | $6.50 | $65.00 |
One document is about 3,000 tokens in and 400 out, at list prices. Try your own numbers in the cost calculator.
Who should pick something else
- For scanned images rather than text, check that the model reads images on its model page.
- For millions of documents a month, look at the open-weights picks and run them yourself.
Questions
Can AI read invoices reliably?
For clear, typed documents, good models are very reliable. Check a sample by hand before you trust any model at volume.
Which is cheapest?
The budget pick below is the cheapest model that passes our bar for this job.
Changes to this guide
- 25 Oct 2026: First version, on provisional scores from Arena and Epoch AI.