Chinese artificial intelligence startup DeepSeek has released an updated version of its AI model, DeepSeek-V4.1-Flash-1017, with improved benchmark performance, faster task completion and shorter outputs as it seeks to enhance efficiency and expand access to its tools.
The company announced on Saturday that the model was available through its website, mobile application and application programming interface (API), according to posts on its official X account.
DeepSeek said the update delivered a “Pareto improvement” in performance and cost, indicating gains in capabilities and efficiency without a corresponding increase in expense.
The model recorded a score of 77.0 on the DeepSWE v1.1 benchmark, up from 74.2 for its previous version. Its score on Automation-Bench V1.0.6 also improved to 60.5 from 54.8, according to a benchmark comparison shared alongside the announcement.
DeepSeek said the updated model had been optimised for DeepSeek Harness, Tencent CodeBuddy and WorkBuddy, among other tools.
It also highlighted the model’s use of reinforcement learning with self-feedback to improve end-to-end project development, including building games and creating 3D models from scratch.
In addition to the model update, DeepSeek opened its Search and Fetch APIs to all users. The services are powered by the company’s in-house search engine.
Each Search or Fetch API endpoint costs US$0.75 per 1,000 calls, the company said.
