Fine-grained visual reasoning on MME Realworld Lite
55.8Avg@1MAPO
Evaluation results
| Method | Links | |
|---|---|---|
| MAPObackbone=Ovis2.5-9B2026.04 | 55.8 | |
| Ovis2.5-9B + GRPOtraining=GRPO2026.04 | 55.5 | |
| GPT-52026.04 | 55.3 | |
| Thymeresults_source=original paper2026.04 | 55.2 | |
| Ovis2.5-9B + GSPOtraining=GSPO2026.04 | 54.9 | |
| Ovis2.5-9B + DAPOtraining=DAPO2026.04 | 50.3 | |
| Mini-o3reproduced_by_authors=true2026.04 | 49.4 | |
| Gemini 2.5 Pro2026.04 | 49.2 | |
| DeepEyesreproduced_by_authors=true2026.04 | 48.4 | |
| Ovis2.5-9B + Coldstart SFTtraining=Supervised Fine-Tuning2026.04 | 47.9 | |
| Ovis2.5-9B2026.04 | 46.1 | |
| Ovis2.5-9B + PPOtraining=PPO2026.04 | 43 |