Visual Document Understanding on MMLongBenchDoc-C
58.2AccuracySynthetic Reasoning
Evaluation Results
| Method | Links | |
|---|---|---|
| Synthetic ReasoningModel family=Qwen3 VL, Scale=32B, Strategy=Synthetic Reasoning2026.03 | 58.2 | |
| Plain DistillationModel family=Qwen3 VL, Strategy=Plain Distillation2026.03 | 57.3 | |
| LongPOModel family=Qwen3 VL, Strategy=LongPO2026.03 | 56.4 | |
| Qwen3 VL 235B A22B InstructModel family=Qwen3 VL, Scale=235B A22B, Strategy=Instruct2026.03 | 56.2 | |
| No-thinkModel family=Qwen3 VL, Strategy=No-think2026.03 | 54.5 | |
| Qwen3 VL 32B InstructModel family=Qwen3 VL, Scale=32B, Strategy=Instruct2026.03 | 53.8 | |
| Synthetic ReasoningModel family=Mistral, Scale=24B, Strategy=Synthetic Reasoning2026.03 | 49.3 | |
| Plain DistillationModel family=Mistral, Strategy=Plain Distillation2026.03 | 47.4 | |
| Qwen Thinking TracesModel family=Mistral, Strategy=Thinking Traces2026.03 | 45.5 | |
| No-thinkModel family=Mistral, Strategy=No-think2026.03 | 44.3 | |
| Mistral 3.1 Small 24BModel family=Mistral, Scale=24B2026.03 | 41.4 |