DeepSeek Ships V4-Flash-Vision-Exp, Claiming Multimodal Agent Performance Near Opus 4.8

DeepSeek (X)via @deepseek_ai·high signalXBlueskyLinkedInCopy link

DeepSeek announced on August 21 that deepseek-v4-flash-vision-exp is live on its API platform, an experimental multimodal model that matches V4-Flash on text, agents, reasoning and world knowledge while making what the company calls a major leap on multimodal agent benchmarks, close to Opus-4.8. Images are billed at up to 384 tokens each at standard V4-Flash pricing, with base64 input across Chat Completions, Messages and Responses endpoints. DeepSeek Harness 0.1.1 shipped the same day with out-of-the-box support, so the cheapest credible vision-agent option just changed.