DeepSeek Releases Open-Weight V4.1 Flash With a More Efficient Architecture
2026-09-10 ยท Frontier Models
DeepSeek released V4.1 Flash, a 552-billion-parameter open-weight model that activates a small fraction of its parameters and sharply reduces memory needs.
DeepSeek released V4.1 Flash on September 10 as the smallest model in a new architecture family, publishing its weights under the MIT license. The company reports a mixture-of-experts design with a 552-billion-parameter backbone that activates 8 billion parameters when reading input and 16 billion when generating output, native image understanding, and a key-value cache about a quarter the size of its predecessor's. From September 14, DeepSeek began routing requests for its flagship V4-Pro to V4.1 Flash.
The release matters beyond benchmarks. Efficiency lowers the resources needed to deploy capable models, and open weights let anyone download, modify, and run them, outside any developer's access controls.
Benchmark comparisons are company-reported and not independently verified.