大模型网摘V4.1-Flash-1017 Arrives With Open Search & Fetch APIs: Key Facts
On October 17, 2026, DeepSeek released DeepSeek-V4.1-Flash-1017 and opened its in-house search engine to all developers via new Search & Fetch APIs. This roundup is based on the official API docs announcement and release notes.
Model update: faster, shorter, cheaper
Compared with V4.1-Flash shipped on September 10, the company calls the update a Pareto improvement in performance and cost: stronger benchmarks, faster task completion, and shorter outputs. The model is live on the website, app, and API, with open weights on Hugging Face and ModelScope. See our RLSF deep dive for the technical angle.
DeepSWE: 77% pass rate at roughly $0.20 per task
On the official DeepSWE v1.1 benchmark (mini-swe-v2 harness), the new model scores 77.0% Pass@1 at an average cost of about $0.20 per task; a low setting clears 72% for around $0.10. The predecessor scored 74.2% at roughly $0.50. Per The Decoder, rival models such as GPT-6 Astra, Gemini 3.8 Flash, and Claude Opus 5 land between $2 and over $10 per task in the official cost chart. For a side-by-side look, see the agentic coding cost comparison.
Search & Fetch APIs open to everyone
The search engine serving hundreds of millions of daily requests is now exposed as two raw JSON APIs: Search returns structured results for a query, and Fetch converts a single web page to Markdown. Each endpoint costs $0.75 per 1,000 calls; calls made via server-side tools in DeepSeek Harness, Claude Code, and Codex are free for DeepSeek API users until October 31, 16:00 UTC. Full parameters and billing notes are in the API guide.
API pricing and migration notes
The API model name remains deepseek-flash: cache-hit input at $0.003–$0.006 per million tokens, cache-miss input at $0.15–$0.30, and output at $0.60–$1.20 (off-peak rates are half). Context is 1M tokens with 384K max output and a 2,500 concurrency limit. Legacy names deepseek-v4-flash and v4-flash-vision-exp are retired and served by the new model at Flash pricing.
Caveats
These benchmark figures come from official charts; independent third-party reviews were not yet available as of publication, and real-world code quality may differ. For earlier hands-on impressions, see the five-week review; the September launch coverage is here.