DeepSeek-V4.1-Flash-1017 Release Notes

Overview of DeepSeek-V4.1-Flash-1017

DeepSeek-V4.1-Flash-1017 is a new model release that delivers a Pareto improvement in performance and cost, characterized by stronger benchmark results, faster task completion, and more concise output generation. The model is now available across DeepSeek's website, application, and API, with specific optimizations for development environments including DeepSeek Harness, Tencent CodeBuddy, and WorkBuddy. Users are encouraged to upgrade to DeepSeek Harness v0.2.1-rc.1 or later to leverage the full capabilities of the new version.

Capabilities in End-to-End Development

DeepSeek-V4.1-Flash-1017 utilizes reinforcement learning with self-feedback to facilitate the creation of complex, end-to-end projects. This capability allows the model to handle tasks ranging from 3D modeling to the construction of web-based games. Demonstrated use cases include:

  • 3D Modeling: Generation of complex architectural scenes, such as the Forbidden City.
  • Game Development: Creation of web-based racing games and recreations of established titles using engines like Godot.
  • Environmental Simulation: Development of 3D scenes depicting historical evolution, such as the "Valley of Ages" project.

Search and Fetch API Availability

DeepSeek has opened its in-house Search & Fetch APIs to all users. These APIs are currently utilized for hundreds of millions of daily global requests.

  • Pricing: The service is priced at $0.75 per 1,000 calls.
  • Promotional Period: Search & Fetch calls integrated via server-side tools in DeepSeek Harness, Claude Code, and Codex are currently provided at no cost. This promotional period concludes on October 31, 2026, at 16:00 UTC, at which point standard billing will commence.

Open-Source Access

The DeepSeek-V4.1-Flash-1017 model has been released as an open-source artifact. It is currently available for download and integration via the following platforms:

Sources