AI BriefingOvis2.6 80B-A3B Released
·2026.05.13 21:29
Key point
Ovis2.6-80B-A3B has been released as an MoE-based multimodal model.
Details
The latest MLLM in AIDC-AI's Ovis series, Ovis2.6-80B-A3B, has been released. It changes the LLM backbone to a Mixture-of-Experts (MoE) architecture, featuring a combination of 80B total parameters and about 3B active parameters.
- Supports a 64K token context.
- Handles image resolutions up to 2880×2880.
- The Think with Image feature re-invokes visual tools like cropping and rotation during the reasoning process to re-examine complex images.
- Enhanced performance in OCR, document understanding, chart interpretation, and long-document QA.
This is a notable update for teams looking to run large-scale multimodal reasoning at low serving cost.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.