DeepSeek adds vision to V4 Flash at the same token price

On August 21, OpenRouter announced DeepSeek V4 Flash Vision Exp, an experimental model from DeepSeek that adds image input to the V4 Flash family. According to OpenRouter's launch post, image input is offered at V4 Flash pricing, and the model matches V4 Flash 0731 on text agent benchmarks. Read together, the claim is that adding the ability to accept images did not reduce measured text-agent performance. The post does not name the benchmarks in its text, and the match is OpenRouter's report rather than an independent test.