[{"data":1,"prerenderedAt":63},["Reactive",2],{"news-story-news-20261026-cursor-deepseek-v4-1-flash-cursorbench-cost-frontier":3},{"id":4,"title_zh":5,"title_en":6,"summary_zh":7,"summary_en":8,"takeaways_zh":9,"takeaways_en":15,"content_zh":21,"content_en":22,"external_links":23,"source":36,"category":43,"impact_score":44,"views_count":45,"likes_count":46,"related_entities":47,"tags":57,"image_url":62},"news-20261026-cursor-deepseek-v4-1-flash-cursorbench-cost-frontier","Cursor 上线 DeepSeek V4.1 Flash：CursorBench 4.0 四档推理全部落在成本前沿，Max 档 47.7%、每任务 $0.98","Cursor adds DeepSeek V4.1 Flash: all four effort levels sit on the CursorBench 4.0 cost frontier, with Max scoring 47.7% at $0.98 per task","Cursor 官方 10 月 27 日（SGT）宣布 DeepSeek V4.1 Flash 已在 Cursor 中可用，在 Cursor Settings > Models 打开即可。Cursor 称该模型每一档推理强度都落在 CursorBench 成本前沿上：每任务 $0.60 到 $0.98，在这个价位没有模型得分更高。官方 CursorBench 4.0 榜单显示，DeepSeek V4.1 Flash 1017 Max 档 47.7%（$0.98/任务），Medium 46.3%（$0.80），High 45.6%（$0.77），Low 44.9%（$0.60）；同榜最高分 Opus 5.5 Max 为 57.8%，但每任务 $13.43。","Cursor announced on Oct 26 (UTC) that DeepSeek V4.1 Flash is now available in Cursor and can be turned on under Cursor Settings > Models. Cursor says every effort level sits on the CursorBench cost frontier, from $0.60 to $0.98 per task, with no model scoring higher at those prices. The official CursorBench 4.0 leaderboard lists DeepSeek V4.1 Flash 1017 at 47.7% on Max ($0.98 per task), 46.3% on Medium ($0.80), 45.6% on High ($0.77) and 44.9% on Low ($0.60). The top score on the board, Opus 5.5 Max, is 57.8% but costs $13.43 per task.",[10,11,12,13,14],"四档跑分（CursorBench 4.0 官方榜）：Max 47.7%/$0.98、Medium 46.3%/$0.80、High 45.6%/$0.77、Low 44.9%/$0.60，Max 档在全榜排第 17","性价比对比：Max 档 47.7% 与 Sonnet 5.5 High（47.8%，$1.20）几乎持平但便宜约 18%；Opus 5.5 Max 高约 10 个百分点，但每任务 $13.43，约是它的 13.7 倍","同价位对手：Haiku 5.5 Max 48.4%/$1.12、Haiku 5.5 Extra High 44.3%/$0.56，DeepSeek 正好卡在两者之间的价格段","代价是 token 和步数：Max 档平均每任务约 53.7 万 token、187 步，Sonnet 5.5 High 只用约 3.7 万 token、41 步，长任务时延要有心理预期","开启方式：Cursor Settings > Models 里打开；有个反直觉细节，Medium 档（46.3%）比 High 档（45.6%）分数略高、价格只贵 $0.03",[16,17,18,19,20],"Four effort levels on the official CursorBench 4.0 board: Max 47.7% at $0.98, Medium 46.3% at $0.80, High 45.6% at $0.77, Low 44.9% at $0.60; Max ranks 17th overall","Value: Max at 47.7% nearly ties Sonnet 5.5 High (47.8%, $1.20) for about 18% less; Opus 5.5 Max scores about 10 points higher but costs $13.43 per task, roughly 13.7x more","Nearest rivals on price: Haiku 5.5 Max at 48.4% for $1.12 and Haiku 5.5 Extra High at 44.3% for $0.56, bracketing DeepSeek's price range","The trade-off is tokens and steps: Max averages about 537K tokens and 187 steps per task, versus about 37K tokens and 41 steps for Sonnet 5.5 High, so expect longer runs","Turn it on under Cursor Settings > Models; a counterintuitive detail is that Medium (46.3%) scores slightly above High (45.6%) for only $0.03 more","### 核心背景与行业痛点\n在 Cursor 里跑 Agent，最贵的往往不是订阅费，而是每个长任务烧掉的模型额度。前沿闭源模型分数最高，但 Opus 5.5 Max 在 CursorBench 4.0 上每任务要 $13.43，日常改 bug、重构用起来心疼。DeepSeek 10 月 17 日发布的 V4.1-Flash-1017 是 MIT 开源权重模型，API 一直以低价著称。这次 Cursor 官方账号宣布把它接进 Cursor，并用自家 CursorBench 给出了成本与分数的对比，给开发者一个「便宜但够用」的官方选项。\n\n### 架构亮点与底层机制\nCursor 提供 Low、Medium、High、Max 四档推理强度，档位越高，模型思考和调用工具的预算越多。CursorBench 4.0 用真实 Cursor 会话中模糊、跨多文件的任务评测 Agent，4.0 版新增了编辑、重构、排查、意图理解、管理任务和遵循设计等长程问题。所谓「成本前沿」，指在同一每任务成本下没有别的模型得分更高。Cursor 口径是 DeepSeek 四个档位全部落在这条前沿上。\n\n### 权威 Benchmark 与实测跑分对比\n以下数字来自 Cursor 官方 CursorBench 4.0 榜单（厂商口径）。DeepSeek V4.1 Flash 1017：Max 47.7%（$0.98/任务，约 53.7 万 token、187 步），Medium 46.3%（$0.80），High 45.6%（$0.77），Low 44.9%（$0.60）。对比：Opus 5.5 Max 57.8%（$13.43），Sonnet 5.5 High 47.8%（$1.20），Haiku 5.5 Max 48.4%（$1.12），Haiku 5.5 Extra High 44.3%（$0.56）。也就是说它的绝对分数不在第一梯队，优势在于同价位最高分。\n\n### 开发者实战落地与开箱指南\n在 Cursor Settings > Models 中打开 DeepSeek V4.1 Flash 即可使用。建议先用 Medium 档：它比 High 档分数略高、只贵 $0.03。难题再切 Max。要注意它每任务消耗的 token 和步数远多于 Sonnet 5.5，长任务会更慢。对质量要求极高的关键改动，仍可切回 Opus 5.5 等高档模型。","### Background\nFor agent work in Cursor, the real cost is often the model spend on each long task rather than the subscription. Frontier closed models score highest, but Opus 5.5 Max costs $13.43 per task on CursorBench 4.0. DeepSeek released V4.1-Flash-1017 on Oct 17 as MIT-licensed open weights with a low-priced API. Cursor's official account has now added it to Cursor and published a cost-versus-score comparison on its own benchmark, giving developers an official cheap-but-capable option.\n\n### How it works\nCursor offers four effort levels, Low, Medium, High and Max; higher levels give the model more budget for reasoning and tool calls. CursorBench 4.0 evaluates agents on ambiguous, multi-file tasks taken from real Cursor sessions, and version 4.0 added long-horizon problems on editing, refactoring, investigation, intent understanding, managing jobs and design adherence. The cost frontier means no other model scores higher at the same cost per task, and Cursor says all four DeepSeek levels sit on it.\n\n### Benchmarks\nAll figures are from Cursor's official CursorBench 4.0 leaderboard (vendor-reported). DeepSeek V4.1 Flash 1017: Max 47.7% ($0.98 per task, about 537K tokens and 187 steps), Medium 46.3% ($0.80), High 45.6% ($0.77), Low 44.9% ($0.60). For comparison: Opus 5.5 Max 57.8% ($13.43), Sonnet 5.5 High 47.8% ($1.20), Haiku 5.5 Max 48.4% ($1.12), Haiku 5.5 Extra High 44.3% ($0.56). Its absolute score is not top tier; its edge is the best score at its price.\n\n### Getting started\nTurn on DeepSeek V4.1 Flash under Cursor Settings > Models. Start with Medium, which scores slightly above High for only $0.03 more, and move to Max for harder tasks. It uses far more tokens and steps per task than Sonnet 5.5, so long runs will be slower. For critical, high-stakes changes you can still switch back to a top model such as Opus 5.5.",[24,28,32],{"title":25,"url":26,"type":27},"CursorBench 4.0 leaderboard (Cursor)","https://cursor.com/cursorbench","docs",{"title":29,"url":30,"type":31},"DeepSeek-V4.1-Flash-1017 release note","https://api-docs.deepseek.com/news/news261017","blog",{"title":33,"url":34,"type":35},"Hugging Face: deepseek-ai/DeepSeek-V4.1-Flash-1017","https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017","demo",{"platform":37,"author_name":38,"author_handle":39,"author_avatar":40,"tweet_url":41,"published_at":42},"x","Cursor","@cursor_ai","https://pbs.twimg.com/profile_images/1970182748146180096/dhZeXi_X_200x200.jpg","https://x.com/cursor_ai/status/2114839997568405608","2026-10-26T22:02:01.000Z","tooling",86,82,0,{"models":48,"agents":54,"vendors":55},[49,50,51,52,53],"DeepSeek-V4.1-Flash-1017","DeepSeek V4.1 Flash","Opus 5.5","Sonnet 5.5","Haiku 5.5",[38],[38,56],"DeepSeek",[38,56,58,59,60,61],"CursorBench","open weights","cost frontier","coding agent","https://pbs.twimg.com/media/HVlrxQPbkAAK0HD.jpg",1793718861290]