DeepSeek introduces V4.1-Flash with native visual understanding
The September release changes the architecture and API model name, while a later notice preserves access to V4-Pro.
THE STORY IN THREE POINTS
- V4.1-Flash adds native visual input.
- The current API name is deepseek-flash.
- The updated changelog says V4-Pro service will continue.
The announcement
DeepSeek announced V4.1-Flash on September 10, describing a 552-billion-parameter mixture-of-experts model with native visual understanding. Its encoder-decoder design activates different parameter counts for input and output. The API uses deepseek-flash. An updated official changelog says V4-Pro service will continue after September 14, superseding the original launch page’s phaseout language. Benchmark and efficiency comparisons remain company claims.
Editorial context
For an application team, a new architecture matters through observable behavior: whether images are understood correctly, responses arrive sooner, and repeated context becomes cheaper. A useful trial would hold prompts and tools constant while changing only the model. That helps distinguish a model improvement from a better agent setup or a shorter task.
Editorial context: what to watch
Compatibility aliases can conceal a model change. Teams could record the actual model version alongside each evaluation and retain difficult image examples as regression cases. The separate V4-Pro continuation notice also illustrates why migration plans should use the current changelog rather than a launch article alone. This is an operational implication, not an independent performance finding.
Sources: DeepSeek: V4.1-Flash announcement, September 10 ↗ DeepSeek API changelog: updated September 10 entry and V4-Pro continuation ↗
