CodeSOTA · Clickable claim · C-0004Evidence snapshot 2026-08-20
Public evidence statement

Mistral OCR 3's verified OmniDocBench component results differ materially across text, tables, formulae and reading order.

Claim
C-0004
Verdict
supported
Verification
CodeSOTA verified
Evidence
4 attached records
Authors
Kacper Wikiel

What this claim answers.

Hypothesis
H-0005 · Component metrics change OCR decisions

Inspect the rows beneath the sentence.

Current registry values are resolved live when available and compared with the captured claim snapshot. Provenance remains visible when resolution falls back to the snapshot.

leaderboard-result:38

mistral-ocr-3 on OmniDocBench

CodeSOTA verified
Model
mistral-ocr-3
Benchmark
OmniDocBench
Metric
text-edit-distance
Value
0.099

Verified text edit distance.

Registry note: Text block recognition. 90.1% accuracy. Best on academic papers (97.9%), exam papers (92.8%).

CodeSOTA legacy leaderboard resultobserved 2025-12-19unchanged since snapshotOpen registry evidence →Primary source ↗
leaderboard-result:39

mistral-ocr-3 on OmniDocBench

CodeSOTA verified
Model
mistral-ocr-3
Benchmark
OmniDocBench
Metric
table-teds
Value
70.88

Verified table TEDS.

Registry note: Table structure recognition. TEDS Structure: 75.3%. Best on exam papers (88.0%).

CodeSOTA legacy leaderboard resultobserved 2025-12-19unchanged since snapshotOpen registry evidence →Primary source ↗
leaderboard-result:40

mistral-ocr-3 on OmniDocBench

CodeSOTA verified
Model
mistral-ocr-3
Benchmark
OmniDocBench
Metric
formula-edit-distance
Value
0.218

Verified formula edit distance.

Registry note: Display formula recognition. 78.2% accuracy.

CodeSOTA legacy leaderboard resultobserved 2025-12-19unchanged since snapshotOpen registry evidence →Primary source ↗
leaderboard-result:41

mistral-ocr-3 on OmniDocBench

CodeSOTA verified
Model
mistral-ocr-3
Benchmark
OmniDocBench
Metric
reading-order
Value
91.63

Verified reading-order score.

Registry note: Reading order accuracy. 8.4% edit distance error.

CodeSOTA legacy leaderboard resultobserved 2025-12-19unchanged since snapshotOpen registry evidence →Primary source ↗

Verification is not the same as reproduction.

codesota verified

“CodeSOTA verified” means the evidence row and its source have passed the registry’s verification path. It does not, by itself, mean an independent team reran the protocol. Reproduction is recorded separately in the claim verdict and experiment graph.

A stable public object.

Created by
Kacper Wikiel
Created
2026-08-20
Snapshot
2026-08-20