DeepSeek has released V4.1-Flash-1017 through its website, app and API, while also publishing the model weights on Hugging Face and ModelScope. The company says the update completes coding and agent tasks faster than its predecessor, produces shorter outputs and improves the balance between performance and cost.
The model uses reinforcement learning with self-feedback, a training method in which internal confidence signals or generated critiques help score its work. DeepSeek says this improves complete software-building tasks, including games, 3D projects and design workflows. It tuned the model for environments including DeepSeek Harness, Tencent CodeBuddy and WorkBuddy.
In DeepSeek’s own DeepSWE benchmark, the model reached a top score of 77 percent at an average cost just below $0.20 per task. A lower setting exceeded 72 percent for about $0.10. Those figures and comparisons with larger rival models have not yet been independently verified.
DeepSeek is also exposing its internal web search through Search and Fetch APIs priced at $0.75 per 1,000 calls. Existing API customers can use server-side integrations in several coding tools free until October 31. The release gives developers both an open model and retrieval services, but real-world code quality remains to be tested.