What changed
DeepSeek opened an internal test of V4.1-Flash on September 8, introducing a new model structure with native multimodal support. Developers can keep the existing `base_url` and call `deepseek-v4.1-flash-expires-on-0910`; billing is currently the same as for DeepSeek V4-Flash.
Why it matters
The practical shift is unusually simple: adding image-and-text capability may not require changing the integration endpoint. That lowers the friction for developers already using DeepSeek’s API, while the temporary model name signals that this is still a test rather than a settled public release.
Multimodal support has been a weaker part of DeepSeek’s lineup. The new test follows the August 21 launch of V4-Flash-Vision-Exp, an experimental vision model that was reported to improve sharply on tasks requiring visual understanding. The latest version is also being described as faster, more capable and cheaper, but those claims have not been quantified with benchmarks or price figures.
The immediate question is whether the temporary endpoint survives beyond September 10 and whether the reported gains hold up outside early feedback. If it does, DeepSeek’s API users could gain a faster route to handling mixed visual and textual inputs without rebuilding their existing setup.
Comments
No comments yet.