DeepSeek Just Released V4.1-Flash-1017, Scores 77.0 on DeepSWE, Opens Search API
https://www.marktechpost.com/2026/10/17/deepseek-just-released-v4-1-flash-1017-scores-77-0-on-deepswe-opens-search-api/📌 DeepSeek V4.1-Flash 再推新 checkpoint，DeepSWE 衝到 77.0 分
TL;DR：DeepSeek 更新 V4.1-Flash checkpoint 並開放 Search／Fetch API，一組 Key 就能串接 coding agent 與網路搜尋。
同一代模型又出新版本，這次 DeepSeek 不只是換個 checkpoint,還順手把自家的網路搜尋引擎開放給所有開發者用——對正在搭 agent 的工程師來說,這代表少裝一套第三方 search API。
🤔 一個月內的第二次迭代，主打「帕累托改善」
DeepSeek-V4.1-Flash-1017 是 V4.1-Flash 的最新 post-training checkpoint，已經在 DeepSeek 網站、App 與 API 全面上線，服務名稱仍是 deepseek-flash。DeepSeek 表示相較於 9 月發布的 V4.1-Flash，新版本分數更高、任務完成更快、輸出也更短，官方稱之為效能與成本上的「帕累托改善」（Pareto improvement）。
🧩 MoE 架構搭配 RL self-feedback，瞄準端到端專案任務
根據 V4.1-Flash 技術報告，這是一個多模態 Mixture-of-Experts 模型，共 40 層 Transformer，拆分為 20 層的 causal encoder 與 20 層的 decoder。DeepSeek 團隊表示 V4.1-Flash-1017 採用強化學習搭配 self-feedback 來處理端到端的專案型任務,範例涵蓋從遊戲製作到 3D 建模。這個版本針對 DeepSeek Harness、Tencent CodeBuddy 與 WorkBuddy 做了調校,官方建議搭配 DeepSeek Harness v0.2.1-rc.1 以上版本使用。
📊 DeepSWE 領先，但 Claude Opus 5 仍守住部分戰場
以下分數均為 DeepSeek 官方自報數字：Terminal-Bench 4.0 在 Claude Code 環境下測試，DeepSWE 在 Mini-SWE-V2 環境下測試，ZeroBench-main 採用 Pass@5 計分。1017 checkpoint 在 DeepSWE（77.0 分）、Automation-Bench、Agents’ Last Exam 與 Apex Bench 上都取得領先；不過 Claude Opus 5 在 Terminal-Bench 4.0、ProgramBench、NL2Repo 與 HLE 上仍維持優勢。
💡 開放 Search／Fetch API：把網路層也變成基礎設施
這次更新同步開放的 Search API 會回傳排序後的網頁搜尋結果,Fetch API 則能把單一頁面轉成 Markdown 格式回傳。DeepSeek 表示其自建搜尋引擎每天服務數億次請求,這代表開發者不必再額外串接第三方搜尋服務,就能讓 agent 具備即時網路存取能力,等於用同一組 API Key 同時拿到模型推理與網路層。
🎯 怎麼接：相容端點與定價
模型可透過 OpenAI 或 Anthropic 格式的端點以 deepseek-flash 呼叫。計價方面，尖峰時段每 1M cache-miss input tokens 收費 0.30 美元,每 1M output tokens 收費 1.20 美元,離峰時段則是半價。對已經在用 DeepSeek 生態系（Harness、CodeBuddy）的團隊,這是一次低摩擦的升級,也值得評估把獨立的 search API 供應商換成 DeepSeek 原生方案。
🔗 來源
- 標題：DeepSeek Just Released V4.1-Flash-1017, Scores 77.0 on DeepSWE, Opens Search API
- 作者／機構：Asif Razzaq, MarkTechPost
- 連結：https://www.marktechpost.com/2026/10/17/deepseek-just-released-v4-1-flash-1017-scores-77-0-on-deepswe-opens-search-api/
#DeepSeek #LLM #CodingAgent #MixtureOfExperts #SearchAPI #AgenticAI #OpenAPI #SoftwareEngineering #ModelUpdate #AIInfrastructure
原始資料 MarkTechPost · 收集於 2026-10-17
摘要原文
DeepSeek has just released DeepSeek-V4.1-Flash-1017 , a new checkpoint of its V4.1-Flash model. It is live on the DeepSeek website, app and API . DeepSeek claims stronger scores, faster task completion and shorter outputs than the September V4.1-Flash release . The company calls it a Pareto improvement in performance and cost. Alongside it, DeepSeek opened its Search and Fetch APIs to every developer. For agent builders, that means a cheaper coding model plus a first-party web layer behind one API key. DeepSeek-V4.1-Flash-1017 is an updated post-training checkpoint of DeepSeek-V4.1-Flash, served through the deepseek-flash model name. The architecture details below come from the V4.1-Flash technical report . DeepSeek team states that DeepSeek-V4.1-Flash-1017 uses reinforcement learning with self-feedback for end-to-end project work. Its examples run from game creation to 3D modeling. It is tuned for DeepSeek Harness , Tencent CodeBuddy and WorkBuddy. DeepSeek recommends DeepSeek Harness v0.2.1-rc.1 or later. V4.1-Flash is a multimodal Mixture-of-Experts model with 40 Transformer layers. They split into a 20-layer causal encoder and a 20-layer decoder. All scores below are DeepSeek’s own figures . DeepSeek ran Terminal-Bench 4.0 in Claude Code and DeepSWE in Mini-SWE-V2. ZeroBench-main uses Pass@5. The 1017 checkpoint leads on DeepSWE, Automation-Bench, Agents’ Last Exam and Apex Bench. Claude Opus 5 still leads on Terminal-Bench 4.0, ProgramBench, NL2Repo and HLE. The Search API returns ranked web results for a query. The Fetch API returns a single page as Markdown. DeepSeek says its in-house engine serves hundreds of millions of requests daily. Call the model with deepseek-flash through OpenAI or Anthropic-format endpoints. Peak pricing is $0.30 per 1M cache-miss input tokens and $1.20 per 1M output tokens. Off-peak rates are half. Check out the DeepSeek announcement , the Search and Fetch docs , the technical report and the model card . All credit goes to the researcher of this project. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter . Wait! are you on telegram? now you can join us on telegram as well. The post DeepSeek Just Released V4.1-Flash-1017, Scores 77.0 on DeepSWE, Opens Search API appeared first on MarkTechPost .
由 claude-code/sonnet 自動生成