DeepSeek updates V4.1 Flash as US discloses federal agency restrictions
DeepSeek says its updated model improves coding and reasoning, while a newly published US order bars its technology from civilian federal systems.
DeepSeek has released V4.1-Flash-1017, an update that the company says improves reasoning, programming and image understanding. It also says the model finishes tasks more quickly while generating shorter responses. DeepSeek published the downloadable model under an MIT license.
In DeepSeek's evaluations, the update scored 42.3 on Terminal-Bench 4.0, compared with 31.2 for the preceding version. Its ProgramBench result increased from 20.3 to 29.1. These are company-published results, rather than guarantees of performance in a particular business application.
The release coincides with public disclosure of US restrictions on DeepSeek in civilian federal agencies. A Homeland Security notice filed October 19 and published October 20 covers procurement and removal from agency technology systems. Its scope includes affiliated companies and third-party offerings built using DeepSeek technology.
The restrictions make the model's role inside government software relevant beyond purchases directly from DeepSeek. Agencies face requirements to exclude covered products from procurement and remove covered technology already in use. The published notice addresses civilian federal systems and purchasing, rather than imposing a general prohibition on private use.
The day's biggest AI news in one short email, every morning. Plus an instant heads-up when a reset or a big release lands.