DeepSeek V4.1 Flash 1017 tops open-source agentic coding benchmarks at a fraction of rivals' cost
DeepSeek 于 10 月 18 日（周六）前后发布开源模型 DeepSeek-V4.1-Flash-1017，已上线官网、App 和 API。官方将其定位为性能与成本的「帕累托改进」：基准成绩更强、任务完成更快、输出更短。发布后数日内，多个第三方基准与实测几乎一致指向同一结论：这是目前性价比极高的开源模型，多项能力逼近甚至超过头部闭源模型。
已确认
- 编码/智能体：DeepSWE v1.1 上取得 77 分（约 77% Pass@1）登顶，且是最便宜模型；Reddit 用户 Informal-Trouble2183 分享的 AI-Leaderboard.dev Agentic Coding Index（聚合 DeepSWE、Terminal-Bench、SWE-bench Pro 等 11 个基准）也显示其登顶开源榜，得分 77.5。博主 eyishazyer 称其在三项 Agent 编码基准超越 Claude Opus 5（DeepSWE 77.0 vs 74.0、AutomationBench 60.5 等）。Yuchen Jin 称其达 Opus 5 水平而成本低 83 倍。
- 浏览器任务：browseruse 基准长程任务 75/100，与 Opus 5.5 持平，单任务成本 $0.22 vs $2.35（约低 19 倍）。
- 3D/视频生成：BuildingBench 3D 建筑重建 81.9 分（上版 72.9），跻身前五，成本约为 Fable 5.1 的 1/11、Sonnet 的约 1/15；Bowser 3D 测试单次 $0.15 vs Opus 5.5 的 $3.85。博主 luckeyfaraday 的视频生成对比中，DeepSeek 花 $0.78、3 分钟完成火箭发射场景，GPT 6.1 Sol 花费 $6.10、耗时 5 分钟。teortaxesTex 转述的建房间实测中 DeepSeek 花约 $0.21、22 分钟，Opus 5.5 花 $2.20、15 分钟，成果相近。Alchain 实测其搭配已开源的 huashu-motion skill 做动效「仅次于 Claude」。
- 搜索：第三方评测方 p0 称其闯入搜索能力榜前十，为开源模型首次，且是其测过的最强表现之一。
- 一手反馈与采用：YouTuber Sentdex 称用下来非常不错；WorldofAI 用自建基准 woaibench.ai 全面实测编码、3D 生成与定价。Nebius 工程师自建 10 个真实工作任务基准（每任务 3 次、隐藏测试校验）全部通过，high 档单任务不到 5 美分。Alex Banks 引用数据称其在 Vercel 上占近一半 token 量，但花费仅 2.6%，以此说明「花费不等于使用量」。Bindu Reddy 称其为「目前世界上最好也最便宜的开源模型」，击败 Muse Wonder、Haiku、Luna 等对手。
尚未确认
- 多条帖子中的对比模型名称（如 Opus 5/5.5、GPT 6.1 Sol、Fable 5.1 等）与具体分数均转自第三方榜单或博主实测，DeepSeek 官方未逐一背书；teortaxesTex 也提过该版本模型标签标注存在疑问。
- Noahbench 给出的评价是「便宜、慢、但相当不错」，排名第五；teortaxesTex 推测耗时偏长可能与「兔子洞」式推理问题有关，属个人猜测。
为什么重要
- 该模型再次冲击「开源追不上闭源」的判断：发布一周内即在编码、浏览器、3D、搜索等多个方向逼近或超越头部闭源模型，且成本普遍低一个数量级以上，可能显著改变智能体应用的单位经济模型。
2026-10-17 ~ 2026-10-19 · 31 related posts
- Episode 1: DeepSeek-V4.1-Flash-1017 Quietly Appears on Hugging Face(2026-10-17, 3 posts)
- Episode 2: DeepSeek API Silently Swapped to Suspected V4.1-Flash-1017 Snapshot(2026-10-17, 6 posts)
- Episode 3: DeepSeek Open-Sources V4.1-Flash-1017, Matching Opus 5 on Multiple Agent Benchmarks(2026-10-17, 48 posts)
- Episode 4: DeepSeek V4.1 Flash 1017 tops open-source agentic coding benchmarks at a fraction of rivals' cost(2026-10-17, 31 posts)
- Episode 5: mrfakename releases abliterated DeepSeek V4.1 Flash and GGUF quants(2026-10-18, 5 posts)
- Episode 6: Unsloth quietly posts DeepSeek-V4.1-Flash GGUF quantization on Hugging Face(2026-10-18, 2 posts)
Primary sources
- DeepSeek V4.1 Flash becomes first open model to crack top 10 on search leaderboard — gaganghotra_ ·
- DeepSeek V4.1 Flash Tops Open-Weight Models on Agentic Coding Index at 77.5 — Informal-Trouble2183 ·
- DeepSeek V4.1 Flash 1017 aces 10-task coding benchmark at under 5 cents per task — Fabulous_Pollution10 ·
- DeepSeek V4.1F-1017 vs V4.1 hands-on: noticeably faster generation — teortaxesTex · 2026-10-17
- Hands-on: DeepSeek V4.1 Flash handles game VFX prompts but falls short of Opus — op7418 · 2026-10-17
- [source] DeepSeek V4.1 Flash Tops Open-Weight Models on Agentic Coding Index at 77.5 — Informal-Trouble2183 · 2026-10-17
- DeepSeek V4.1-Flash-1017 claimed to hit Opus 5-level coding scores at 83x lower cost — Yuchenj_UW · 2026-10-18
- DeepSWE benchmark: DeepSeek V41 Flash-1017 tops chart at 77% pass rate with lowest cost — Immediate_Simple_217 · 2026-10-18
- DeepSeek video model test: 8x cheaper and faster than GPT 6.1 Sol — gaganghotra_ · 2026-10-18
- [source] DeepSeek V4.1 Flash 1017 aces 10-task coding benchmark at under 5 cents per task — Fabulous_Pollution10 · 2026-10-18
- DeepSeek V4.1-Flash-1017 outscores Claude Opus 5 on three agentic coding benchmarks — eyishazyer · 2026-10-18
- 500M tokens for $4: hands-on with DeepSeek V4.1 Flash inside Claude Code, 9 prompts released — AlchainHust · 2026-10-18
- DeepSeek V4.1 Flash matches Opus 5.5 on 3D test at $0.15 vs $3.85 per run — SumitGup · 2026-10-18
- DeepSeek V4.1 Flash 1017 ranks 5th on new Noahbench eval, "cheap, slow, decent" — teortaxesTex · 2026-10-18
- DeepSeek V4.1 Flash becomes everyone's 3D studio at up to 26x lower cost — 量子位 · 2026-10-18
- DeepSeek drops 'best and cheapest' open-source model, tops leaderboard per Bindu Reddy — bindureddy · 2026-10-18
- DeepSeek V4.1 Flash enters BuildingBench top five at 1/15 of Sonnet's cost — Lianhuiq · 2026-10-18
- DeepSeek V4.1 Flash jumps on long-run agentic tasks but regresses on constraints — teortaxesTex · 2026-10-18
- DeepSeek V4.1 Flash Fully Tested: Coding, 3D Generation and Ultra-Low API Pricing Benchmarked — WorldofAI · 2026-10-18
- DeepSeek V4.1 Flash 1017 matches Opus 5.5 on long-horizon browser use at 19x lower cost — zizhpan · 2026-10-18
- DeepSeek-V4.1-Flash video test: 225 min, $2.4, ~80% of Opus5.5 quality — AlchainHust · 2026-10-18
- DeepSeek's new model nearly matches Claude at motion graphics via huashu-motion skill — AlchainHust · 2026-10-18
- DeepSeek V4.1 Flash serves nearly half of Vercel tokens at just 2.6% of spend — thealexbanks · 2026-10-18
- Same agent build task: DeepSeek costs $0.21 vs Opus 5.5's $2.20 — teortaxesTex · 2026-10-19
- [source] DeepSeek V4.1 Flash becomes first open model to crack top 10 on search leaderboard — gaganghotra_ · 2026-10-19
- DeepSeek V4.1 Flash matches Astra on 3D, but vision quality still lags rivals — bookwormengr · 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
- DeepSeek V4.1 Flash 1017 beats larger rivals on agentic tasks while slashing per-task costs — yogthos · 2026-10-19
- Sentdex: DeepSeek v4.1 Flash 1017 looking very good so far — Sentdex · 2026-10-19
- DeepSeek's V4.1-Flash-1017: 500M tokens for ~$4, sparking heavy testing — sven_ai · 2026-10-19
- DeepSeek V4.1 Flash sparks buzz: 81.9 score at ~1/11 the cost, third-party evals pending — teortaxesTex · 2026-10-19