DeepSeek bolts vision onto V4 Flash
DeepSeek released an experimental multimodal version of V4 Flash that accepts visual prompts, claiming performance nearing Anthropic's Opus 4.8 on multimodal agentic tests, per Bloomberg. The API docs went live today and the Hacker News thread cleared 430 points within hours.
- V4 Flash is the smaller of the V4 pair — 284B total / 13B active parameters, MIT-licensed weights, 1M-token context
- Bloomberg frames the release as a direct benchmark challenge to Anthropic's frontier multimodal models
- The vision variant ships as an '-exp' experimental API release, ahead of any weights drop