V4.1 Flash 1017 ships with open weights; Search & Fetch at $0.75 per 1,000 calls
Roundup of official details: higher benchmarks and shorter outputs, RL-trained full-project generation, Search & Fetch APIs at $0.75 per 1,000 calls, free in Harness/Claude Code/Codex until Oct 31.
DeepSeek just dropped V4.1-Flash-1017, and the weights are open-source on day one. What stands out: Performance: Higher benchmark scores, faster task completion, and shorter outputs than the last version. DeepSeek says it's better and cheaper at the same time 📈 Full-project coding: Trained with RL and self-feedback to build whole projects from scratch, like a web racing game, a 3D Forbidden City, and a Portal 2 recreation in Godot 🎮 Search & Fetch APIs: Now open to everyone at $0.75 per 1,000 calls per endpoint. These calls are free through DeepSeek Harness, Claude Code, and Codex until Oct 31 💰 Availability: Live now on DeepSeek's website, app, and API, with weights on Hugging Face and ModelScope 🔓