In the open-source large model field, another technological breakthrough has occurred! According to the latest evaluation list released by Arena, the latest model iteration version DeepSeek-V4.1-Flash-20261017 (Max) launched by the DeepSeek team has officially topped the Code Arena: WebDev list, winning the title of “Top Open-Source Model”.
It is worth noting that in this WebDev cutting-edge code and web page development benchmark test, DeepSeek-V4.1-Flash-1017 not only surpassed the previously notable Kimi K3 (Max), but also achieved a substantial breakthrough in inference cost—its invocation overhead is only about one tenth of that of the latter. Between performance and cost-effectiveness, it has once again updated the industry-recognized Pareto optimal boundary.
In terms of overall ranking, this model ranked 12th on the total list with a score of 1670. In this high-tier category, it not only leads over the closed-source leader model Claude Opus5 (High), but the gap with the top model GPT-6Sol (Max) has also narrowed to less than 10 points.
According to the official introduction, the version 1017 released this time is a deeply optimized iteration. Compared with the previous version (20260910) released a month ago, the ranking score has increased by 52 points. While maintaining the extremely low pricing strategy (0.3 US dollars per million input tokens and 1.2 US dollars per output token), the model has further accelerated the completion of complex tasks and significantly reduced redundant outputs, helping developers save computing power and time costs in actual usage.
Currently, DeepSeek-V4.1-Flash-1017 has been synchronized and available on the DeepSeek web platform, mobile App, and API interface. For the developer ecosystem, the official team has conducted specialized adaptation and optimization for several mainstream AI programming and Agent assistance tools such as DeepSeek Harness, Tencent CodeBuddy, and WorkBuddy. Developers only need to upgrade the framework to v0.2.1-rc.1 or later to seamlessly unlock the highest performance experience of this model.