China's DeepSeek has pushed an update to its fastest model. DeepSeek V4.1 Flash is now DeepSeek-V4.1-Flash-1017, and the official web and app models have been updated too, the company said in its API change log.
The model weights were posted on Hugging Face on Oct. 17 (UTC), where the repository had more than 10,000 downloads by Sunday.
What changed
- Benchmarks: DeepSeek claims substantial improvements over the previous V4.1 across benchmarks including HLE (text subset), Terminal-Bench 4.0 and DeepSWE.
- Cost and speed: it says the model cuts task runtime and output length, so tasks finish faster and cheaper, a "performance–cost Pareto" improvement.
- Price: API calling and pricing are unchanged. Developers use the model name
deepseek-flash.
DeepSeek says it tuned the update for its own DeepSeek Harness and for Tencent's CodeBuddy and WorkBuddy, and recommends DeepSeek Harness 0.2.1-rc.1 or later.
Background
V4.1 Flash is the smallest model in DeepSeek's new architecture family, with native multimodal visual understanding, according to the change log. The update is DeepSeek's claim, not independently benchmarked here.