DeepSeek ships V4.1-Flash-1017 with Pareto gains, now on Ollama with zero-retention US hosting
gaganghotra_ · x · 2026-10-18
DeepSeek officially released DeepSeek-V4.1-Flash-1017, claiming stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in both performance and cost. It's live on the website, app, and API.
- Optimized for DeepSeek Harness, Tencent CodeBuddy, WorkBuddy and more; DeepSeek Harness v0.2.1-rc.1+ recommended.
- Ollama added the model the same day with US-based hosting and zero data retention, highlighting strong agentic-task performance at lower per-task cost, runnable via Ollama Cloud in Claude Code and OpenCode.
More from coding & agent
- Don't run bot teams on scheduled wake-ups: an event-driven agent architecture — b_renat · 2026-10-19
- Claude one-shot rebuilds a structural integrity physics demo from a video — erwincoumans · 2026-10-19
- 300 parallel agents plus a 100ms cheap verifier compress weeks of analyst work into one run — CurieuxExplorer · 2026-10-19
- Local Qwen vision model builds a 3D room game in 26 minutes on a single prompt — Beneficial-Ad-8127 · 2026-10-19
- Building a vast Unreal Engine world from scratch with Opus, says AI game dev is accelerating — sven_ai · 2026-10-19
- DeepSeek V4.1-Flash lands #2 on Terminal-Bench 4.0 at 1/16th the price of Opus 5 — zainhas · 2026-10-19