DeepSeek updates V4.1 Flash
A new build of DeepSeek's fast model. See the change log for what is new.
DeepSeek has upgraded V4.1 Flash to a version called DeepSeek-V4.1-Flash-1017, now used in its web app, mobile app and API. API users keep the same model name, deepseek-flash, and the same prices. DeepSeek reports higher scores than the previous DeepSeek V4.1 on tests such as DeepSWE v1.1 and Terminal-Bench 4.0.
DeepSeek says the new version also finishes tasks faster and with shorter output, so the same job costs less. In the same update the company opened its in-house search engine to API users, saying it filters SEO spam and low-quality AI-generated content.
Existing integrations need no changes, since the model name and calling method stay the same. DeepSeek says it tuned the new version for DeepSeek Harness, Tencent CodeBuddy, WorkBuddy and other platforms, and recommends DeepSeek Harness 0.2.1-rc.1 or later for the best results.