DeepSeek V4.1 Flash 1017: Wish-Style Programming & Open Search API
DeepSeek releases V4.1 Flash 1017 with better performance and lower cost, introduces wish-style programming via self-feedback RL for building 3D projects and games from minimal prompts, opens self-developed Search and Fetch APIs at 5 yuan per 1000 calls, and open-sources model weights on HuggingFace and ModelScope.
Performance and Cost Improvements
DeepSeek V4.1 Flash has been updated to version 1017 across web, app, and API, with model weights open-sourced. Benchmarks show comprehensive gains while task runtime and output length decrease. The company describes this as a Pareto improvement in performance‑cost trade‑offs. Special optimizations target DeepSeek Harness, Tencent CodeBuddy, and WorkBuddy frameworks; Harness users should upgrade to 0.2.1‑rc.1 or later.
Wish‑Style Programming
The core highlight is "wish‑style programming" (许愿式编程). Through self‑feedback reinforcement learning (RL) training, the new model can construct complete projects from scratch in game development and 3D modeling with very few prompts and interaction rounds. Four official demos, all built inside the DeepSeek Harness framework, illustrate the capability:
Forbidden City 3D modeling showcase
Web‑based racing game
Portal 2 replica using Godot engine
Valley of Ages: dynamic 3D timeline from prehistory to modern era
In short, describe the desired effect and the model produces a runnable project.
Self‑Developed Search API Opened
DeepSeek's search capability has been powered by a proprietary engine handling hundreds of millions of daily requests worldwide. The Search and Fetch endpoints are now publicly available, priced at 5 CNY per 1,000 calls. API documentation is provided in the official guide. Note that when using DeepSeek API via DeepSeek Harness, Claude Code, or Codex, server‑side tool calls already route through Search/Fetch; this usage is free until 2026‑11‑01 00:00 Beijing time, after which per‑call billing begins.
Model Open‑Source Release
Model weights are published on two platforms:
HuggingFace: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017
ModelScope: https://modelscope.cn/models/deepseek-ai/DeepSeek-V4.1-Flash-1017
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
AI Engineering
Focused on cutting‑edge product and technology information and practical experience sharing in the AI field (large models, MLOps/LLMOps, AI application development, AI infrastructure).
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.