Overview
- DeepSeek opened a limited internal test for DeepSeek V4.1 Flash on Tuesday, and the temporary model endpoint is callable as deepseek-v4.1-flash-expires-on-0910.
- The test endpoint uses the existing base_url, charges at the same rate as V4 Flash, and limits each account to 20 concurrent requests.
- DeepSeek describes V4.1 Flash as a new model structure with native multimodal support, which would integrate image plus text capability into the base model rather than keeping vision as a separate experiment.
- The company has not published technical details, parameter counts, or benchmarks for V4.1 Flash, and the model is not yet listed in public API docs or the Change Log.
- Early hands-on checks reported faster generation and improved responses on some writing and UI tasks, but results were mixed on larger integrated problems, so independent testing will be needed to judge whether Flash can match Pro and change developer cost choices.