DeepSeek releases updated V4.1-Flash model it says completes tasks faster

DeepSeek's V4.1-Flash-1017 update aims to improve efficiency, claiming reduced task execution time and output length, though the company did not specify the extent of these reductions. This incremental release follows the original V4.1-Flash, which was noted for its competitive pricing, with off-peak API rates starting at 15 cents per million uncached input tokens, a factor Bloomberg reported led to share declines for Chinese rivals MiniMax Group Inc. and Alibaba Group Holding Ltd. in Hong Kong.

The updated model shows a modest improvement in DeepSeek's internal DeepSWE v1.1 software engineering test, resolving 77% of tasks compared to 74.2% for the September version. While this places it slightly above Anthropic PBC's Claude Opus 5 on this specific metric, DeepSeek's own published benchmarks still show Anthropic's model leading in seven out of eleven categories, with a notable margin on Humanity's Last Exam.

DeepSeek has also expanded its offerings for agent developers. The V4.1-Flash-1017 was specifically tuned for agent software, including its DeepSeek Harness tool and Tencent's CodeBuddy coding assistant and WorkBuddy desktop agent. Additionally, DeepSeek opened its in-house search engine to external developers through a Search API, returning up to 50 ranked web results per query, alongside a Fetch API for converting web pages to Markdown. These APIs are priced at 75 cents per 1,000 successful requests, with free calls until November 1 for those made through DeepSeek Harness, Anthropic's Claude Code, or OpenAI Group PBC’s Codex.