**[DeepSeek Brings AI Programming into the "Wishful Era": Creating 3D Games from Scratch, DeepSWE Outperforms Opus 5]** Beating AI News Flash: DeepSeek has upgraded V4.1 Flash to version 1017, with the web version, app, and API now fully synchronized. The new model weights have also been open-sourced under the MIT license. In official tests, DeepSWE v1.1's pass rate for software engineering tasks increased from 74.2% to 77.0%, surpassing Claude Opus 5's 74.0%. Terminal-Bench 4.0 improved from 31.2% to 42.3%, though it still falls short of Opus 5's 51.8%. According to the officially released cost curve, achieving a 77% pass rate configuration costs an average of approximately ¥1.3 per question. However, the comparison is against Opus 5, not the newer Opus 5.5. The new version emphasizes "wishful programming," where users provide simple requirements, and the AI autonomously plans and completes the entire project. DeepSeek leverages self-feedback reinforcement learning to enhance these capabilities and has showcased projects such as a 3D scene of the Forbidden City, a web-based racing game, and a recreation of *Portal 2* built on the Godot engine. The official statement notes that the new model has reduced task completion time and output length, while also optimizing for DeepSeek Harness, Tencent CodeBuddy, and WorkBuddy. Additionally, DeepSeek has, for the first time, opened its proprietary search engine's Search and Fetch APIs to developers. Developers can enable agents to directly search web pages and extract content. Both interfaces are billed per successful request, costing ¥5 per 1,000 requests. Search requests triggered via Claude Code or Codex's built-in web search functionality will have their API call fees waived for the first two weeks. [Original Link]