AitopkitDeepSeek released DeepSeek-V4.1-Flash-1017 on October 17, an updated post-training checkpoint of the V4.1-Flash model it launched in September. It is live on DeepSeek's website, app and API under the same deepseek-flash model name, with MIT-licensed weights on Hugging Face and ModelScope.
DeepSeek calls the update "a Pareto improvement in performance and cost": it claims stronger scores on all 11 benchmarks it reports, plus faster task completion and shorter outputs — at unchanged pricing ($0.15 per million input tokens, $0.60 per million output tokens). The 1-million-token context window stays the same. One important caveat: every score is DeepSeek's own, produced on test harnesses it chose — none has been independently reproduced.
The standout claim is 77.0 on DeepSWE v1.1, a coding benchmark where DeepSeek's table puts it ahead of Claude Opus 5 (74.0) and GPT-5.6 Sol (73.0). The picture is less flattering elsewhere in its own comparison: Opus 5 still leads on Humanity's Last Exam and Terminal-Bench 4.0, so "on par with Opus 5" chatter is not supported by the published numbers. The checkpoint is tuned for agent harnesses — DeepSeek's own Harness, Tencent's CodeBuddy and WorkBuddy — with reinforcement learning and self-feedback aimed at end-to-end project building, from games to 3D models.
Arguably the bigger story is infrastructure. DeepSeek opened its Search and Fetch APIs to all developers, powered by an in-house search engine it says serves hundreds of millions of requests a day. Each endpoint costs $0.75 per 1,000 calls, and calls made through server-side tools in DeepSeek Harness, Claude Code and Codex stay free for DeepSeek API users until per-request billing starts on November 1. For agent builders, that means a cheap coding model plus a first-party web layer behind a single API key — putting DeepSeek in direct competition with dedicated search-API providers for agentic workloads.
The release lands as DeepSeek's latest funding round is reportedly growing to $12–15 billion at a valuation of about $75 billion, ahead of a planned Shanghai IPO, according to CNBC. Cheap, open weights with competitive coding scores is a real result — the benchmark tables are marketing until third parties reproduce them.