Deepseek releases experimental Flash vision model that rivals Opus 4.8 on agent benchmarks

Insight summary

• •Deepseek released V4-Flash-Vision-Exp, an experimental multimodal model adding image understanding to text capabilities.
• •The model nearly matches Opus 4.8 performance on Deepseek's internal multimodal agent benchmarks.
• •It supports image formats like JPEG, PNG, GIF, and WebP, determining format from file content.
• •Designed for agent-based applications, it integrates visual understanding with tool use and works with OpenAI and Anthropic APIs.
• •Developers can send images via Base64 encoding, public URLs (up to 32 MiB), or a free Files API (up to 64 MiB).
• •Image inputs can be downscaled for token savings, with token cost capped per image; max 600 images per request with size limits based on image count.
• •Pricing aligns with the V4-Flash model rates.