Deepseek Launches Multimodal Model V4FlashVisionEXP

歸藏的AI工具箱 · wechat · 2026-08-21

Deepseek has released a standalone multimodal model, DeepseekV4FlashVisionEXP. It claims capabilities approaching Opus 4.8 on multimodal agent tasks, while maintaining the same text capabilities and pricing as DeepseekV4Flash. DeepseekHarness has added support, and a free File API was launched simultaneously, allowing image reuse via ID to avoid re-uploading.

Related event: DeepSeek Launches V4-Flash-Vision-Exp, Nearing Opus 4.8 on Multimodal Agent Benchmarks(10 posts)→

Original post →

More from Multimodal

Multimodal channel →