DeepSeek-V4.1-Flash-1017 tops Arena Code Arena WebDev and Agent Arena open rankings
According to 智东西, Arena announced on October 19 that DeepSeek-V4.1-Flash-1017 (Max) became the highest-ranked open model on its Code Arena: WebDev leaderboard. The model scored 1670 points, ahead of Kimi K3 (Max) and Mimo-V2.6 Flash and also ahead of closed models including Muse Spark 1.3, Claude Opus 5 and Grok 4.7, while finishing 12 points behind GPT-6 Sol (Max). IT之家 reported on October 20 that Arena also ranked the model first among open models and 11th overall on Agent Arena; IT之家 cited 7,498 real usage sessions and a 5.13% net improvement, while an Arena post cited a 5.1% net improvement across more than 7,500 sessions. IT之家 put median cost per task at about $0.18–0.19, while the Arena post cited $0.20; Arena listed the model on its performance-versus-cost Pareto frontier. The model was released on October 17. DeepSeek says it supports “wish-style programming” that can generate games and 3D models in one click, and 智东西 reported that its average cost per task on the DeepSWE v1.1 software engineering test fell about 60% from the previous version.
The reports frame the result as a cost-performance competitive development: Arena said the model reshaped its Agent Arena Pareto frontier, with models above it in performance being proprietary and more expensive, while 智东西 said the update strengthened DeepSeek's competitiveness between coding ability and cost.
智东西 编译 | 杨京丽 编辑 | 李水青
智东西10月19日消息，今日，AI模型评测平台Arena宣布，DeepSeek-V4.1-Flash-1017（Max）超过Kimi K3（Max），成为Code Arena: WebDev榜单排名最高的开源模型。
截至目前，DeepSeek-V4.1-Flash-1017在该榜单上得分为1670分，超过了Kimi K3（Max）、Mimo-V2.6 Flash等开源模型，也超过了Muse Spark 1.3（Max）、Claude Opus 5（High）、Grok 4.7（xHigh）等闭源模型，仅落后GPT-6 Sol（Max）12分。
该模型于10月17日推出，在软件工程能力DeepSWE v1.1测试中，其得分超GPT-6 Astra、GPT-6.1 Sol、Claude Opus 5等模型，单项任务平均费用较上一版本下降约六成。DeepSeek称其模型支持“许愿式编程”，可一键生成游戏、3D建模等。网友实测该模型生成速度比GPT-6.1超快模式还要快，甚至有开发者喊话：这得改名叫“V4.2-Pro”了。
#DeepSeek#DeepSeek-V4.1-Flash-1017#Arena#Kimi K3#GPT-6 Sol#Claude Opus 5#Agent Arena#Code Arena