Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co. Ltd. has launched an updated version of its V4.1-Flash model, designated as V4.1-Flash-1017. This release follows the initial launch of the V4.1-Flash model, which occurred five weeks prior. According to the company, the update is designed to reduce task execution time and output length, although specific performance metrics regarding the speed increase were not disclosed.
In internal testing on the DeepSWE v1.1 software engineering benchmark, the updated model achieved a 77% success rate, compared to 74.2% for the previous version. The model is specifically tuned for agent software, including the DeepSeek Harness tool and Tencent Holdings Ltd. products, such as the CodeBuddy coding assistant and the WorkBuddy desktop agent.
DeepSeek has also introduced a new Search API, providing up to 50 ranked web results per query, and a Fetch API for converting web pages into Markdown. Both services are priced at $0.75 per 1,000 successful requests. Calls made through DeepSeek Harness, Anthropic’s Claude Code, or OpenAI’s Codex remain free until November 1.
The update follows a five-hour API outage on October 15. In response, DeepSeek issued credits to users based on their previous week's spending and announced plans to eliminate single points of dependency in its infrastructure. The new model is currently available via web and mobile applications, with API access provided through the existing deepseek-flash model ID.
