Codesota · Models · Mistral OCR 3Mistral6 results · 2 benchmarks
Model card

Mistral OCR 3.

MistralapiVision-Language Model1 current SOTA

Latest Mistral OCR (Dec 2025). 74% win rate vs OCR 2. Claims 94.9% accuracy. Markdown + HTML table output. $1/1000 pages with batch API.

§ 02 · Benchmarks

Every benchmark Mistral OCR 3 has a recorded score for.

#BenchmarkArea · TaskMetricValueRankDateSource
01OmniDocBenchComputer Vision · Document Parsingreading-order91.6%#1/2source ↗
02OmniDocBenchComputer Vision · Document Parsingtext-edit-distance0.1%#2/3source ↗
03OmniDocBenchComputer Vision · Document Parsingformula-edit-distance0.2%#2/2source ↗
04OmniDocBenchComputer Vision · Document Parsingtable-teds70.9%#3/4source ↗
05olmOCR-BenchComputer Vision · Document Parsingpass-rate78.0%#11/21source ↗
06OmniDocBenchComputer Vision · Document Parsingcomposite79.8%#27/34source ↗
Rank column shows this model’s position vs all other models scored on the same benchmark + metric (competitors after the slash). #1 in red means current SOTA. Sorted by rank, then newest result.
Research graph

Research connected to this record.

1 question · 4 claims · 1 experiment

Questions, claims, and experiments linked through CodeSOTA’s public evidence graph.

Explore the research graph →
§ 03 · Strengths by area

Where Mistral OCR 3 actually performs.

Computer Vision
2
benchmarks
avg rank #7.7 · 1 SOTA
§ 05 · Related models

Other Mistral models scored on Codesota.

Mixtral-8x22b
17 results
Mixtral-8x7b
16 results
Mistral-7B-Instruct-v0.2
9 results
Mistral-7B-Instruct-v0.3
9 results
Mistral-Large-Instruct-2407
9 results
Mistral-Nemo-Instruct-2407
9 results
Mistral-Small-24B-Instruct-2501
9 results
Mistral-Small-3.1-24B-Instruct-2503
9 results
§ 06 · Sources & freshness

Where these numbers come from.

codesota-verified
4
results
mistral-announcement
1
result
CodeSOTA (verified)
1
result
5 of 6 rows marked verified.