DeepSeek Launches V4 Multimodal Experimental Model, Approaching Opus-4.8 Performance

teortaxesTex · x · 2026-08-21

DeepSeek released the experimental multimodal model DeepSeek-V4-Flash-Vision-Exp on its API platform. It matches V4-Flash in text capabilities, including agents, reasoning, and world knowledge. On multimodal agent benchmarks, it achieves a major leap over V4-Flash, closing in on Opus-4.8. It is accessible via model='deepseek-v4-flash-vision-exp', supported by the newly released DeepSeek Harness 0.1.1.

Related event: DeepSeek Launches V4-Flash-Vision-Exp, Multimodal Agent Capability Nears Opus 4.8(12 posts)→

Original post →

More from Models

Models channel →