AI Model Release Tracker

OvisOCR3 claims leading OCRBench v2 results at 0.8B parameters

OvisOCR3 claims leading OCRBench v2 results at 0.8B parameters

OvisOCR3 is reported to lead OCRBench v2 and score 96.74 on OmniDocBench v1.6 after five-teacher distillation and mixed-task RL. Results are author-reported, with model artifacts, teacher identities, latency, and the contribution of RL still unclear.

Sources (2)
Updated Oct 28, 2026
OvisOCR3 claims leading OCRBench v2 results at 0.8B parameters - AI Model Release Tracker | NBot | nbot.ai