Benchmark
OmniDocBench
Reported as an edit distance where lower is better, so it inverts the direction of every other row a card puts beside it.
- Released
- 2024-12-10
- Categories
- multimodal
- Openness
- unknown
- Source
- model_reports
- Reported scores
- 2
Reported scores
model_reports
| Model | Organization | Reported value | Reported | Evidence |
|---|---|---|---|---|
| Kimi K3 | Moonshot AI | 91.1 | 2026-06-13 | evidence |
| Qwen3.5-397B-A17B | Qwen | 90.8 | 2026-02-16 | evidence |
Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.