AI Lab
Official DeepSeek announcements — model releases, product launches and research, each one summarized with every source covering it, by AIBriefs. RSS
DeepInfer, the engine behind DeepSeek-V4.1-Flash, ships under Apache 2.0 with a companion MIT-licensed deepseek-recipe token layer; DeepSeek says one cluster serves 155 trillion tokens per day. It supports Nvidia SM100 and Huawei Ascend 950, but omits the KV cache system, P/D disaggregation and load balancing.
The MIT-licensed update to V4.1 Flash cuts task runtime and output length, scoring 42.3 on Terminal-Bench 4.0 (up from 31.2) and 77.0 on DeepSWE. It keeps the 552B backbone activating 8B parameters in prefill and 16B in decode, with a 1M-token context.
DeepSeek released packaged desktop builds of its open-source agent harness for macOS (Apple silicon) and Windows (64-bit), with Linux available via the @deepseek-ai/dsh npm package. The v0.2 preview adds an Optional Bundle architecture, Windows sandbox improvements, async question mode, and keyless web search.