Daily AI news, sourcedOfficial news, first rumor reports and substantive progress. Each entry keeps the information published at that time.
- DeepSeek-V4.1-Flash-1017 is live on Alibaba Cloud's Token Plan
Alibaba Cloud says DeepSeek-V4.1-Flash-1017 is now on its Token Plan, citing 77.0 pass@1 on DeepSWE v1.1, a 1M-token context and 384K output, at the same price as deepseek-v4.1-flash with off-peak discounts.
- DeepSeek upgrades V4.1 Flash to the 1017 snapshot and open-sources its weights
DeepSeek's API change log says V4.1 Flash has been upgraded to DeepSeek-V4.1-Flash-1017, and the official web and app now use the new version too. API calls and pricing are unchanged: the deepseek-flash model name now points to the 1017 snapshot. DeepSeek reports gains over the September V4.1 Flash release, including Terminal-Bench 4.0 at 42.3 (up from 31.2), DeepSWE v1.1 at 77.0 (from 74.2) and HLE without tools at 42.6 (from 36.8), while cutting task runtime and output length; these scores are self-reported by DeepSeek. It says the model is tuned for DeepSeek Harness, Tencent CodeBuddy and WorkBuddy, and recommends DeepSeek Harness 0.2.1-rc.1 or later. DeepSeek says self-feedback reinforcement learning improved the model's ability to build complete projects from scratch, such as games and 3D scenes, and it has released the 1017 weights under the MIT license on Hugging Face and ModelScope.
- Test code in DeepSeek's harness repo hints at DeepSeek-V4-Mini with 1M context
Test code in DeepSeek's public deepseek-harness GitHub repository references a deepseek-v4-mini model (display name DeepSeek-V4-Flash, 1M context window) alongside a deepseek-v4-pro entry, suggesting an upcoming lighter model option in the account system. The references appear in unit-test mocks and DeepSeek has not officially announced the model; DeepSeek-V4-Flash has been available via the official API since July.