DeepSeek launches V4.1-Flash-1017 model with faster task execution
The Chinese AI startup released a minor update to its smallest model, improving software engineering benchmarks and agent compatibility.
Key points
- DeepSeek released V4.1-Flash-1017, a minor update that reduces task execution time and output length.
- The new model achieved 77% on the DeepSWE v1.1 test, up from 74.2% in the previous version.
- The company launched Search and Fetch APIs for developers, with free access until November 1.
Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co. Ltd. released an updated version of its V4.1-Flash model on Hugging Face. The company describes the release as a minor update designed to reduce task execution time and shorten output length, arriving just over five weeks after the original launch.
The new model, identified as DeepSeek-V4.1-Flash-1017, shows improved performance in internal testing. It resolved 77% of tasks on the DeepSWE v1.1 software engineering benchmark, an increase from the 74.2% score achieved by the September version. Anthropic PBC’s Claude Opus 5 holds a 74% score on the same test, according to DeepSeek’s published data.
Performance and Agent Integration
Despite the improvement, Anthropic’s model still leads in seven of the eleven benchmarks DeepSeek published. The margin on Humanity’s Last Exam remains nearly 14 points in favor of the competitor. DeepSeek tuned the updated model specifically for agent software, including its own DeepSeek Harness tool.
The update also supports two products from Tencent Holdings Ltd.: the CodeBuddy coding assistant and WorkBuddy, a desktop agent for office work. WorkBuddy has supported the V4.1-Flash model since DeepSeek released the original version on September 10. The updated model is now live in DeepSeek’s web and mobile applications.
New APIs and Market Reaction
DeepSeek opened its internal search engine to external developers through a new Search API, which returns up to 50 ranked web results per query. A separate Fetch API converts single web pages into Markdown format. Both services cost 75 cents per 1,000 successful requests, but calls made through DeepSeek Harness, Anthropic’s Claude Code, or OpenAI Group PBC’s Codex are free until November 1.
The original model’s low pricing impacted competitors’ stock prices. Off-peak API rates started at 15 cents per million uncached input tokens when the model launched in September. Bloomberg reported that shares of Chinese rivals MiniMax Group Inc. and Alibaba Group Holding Ltd. fell in Hong Kong following that release.
Recent Operational Challenges
The release follows a significant service disruption on October 15. Yicai Global counted the nearly five-hour API outage among DeepSeek’s worst incidents this year. The company apologized and provided users with credits based on their spending from the previous week.
DeepSeek stated it would remove single points of dependency in its systems to prevent similar failures. Some experts quoted by Bloomberg attributed the outage to a shortage of computing capacity. The company has not confirmed whether the new model addresses these infrastructure constraints.
What should you do?
Security and technology professionals should monitor DeepSeek’s API documentation for changes in rate limits or authentication requirements. Developers using the new Search and Fetch APIs should review the pricing structure and free-tier expiration date of November 1.
What to do and how to stay safe: DeepSeek
- Review application logs for any anomalies related to API response times or output length changes after integrating the new model.
- Verify that existing agent software configurations remain compatible with the updated V4.1-Flash-1017 model parameters.
- Monitor vendor announcements for any further infrastructure updates or outage reports from DeepSeek.
- Assess whether the improved benchmark scores impact your organization’s risk assessment for automated code generation tasks.
Step-by-step guide: Mobile Device Security: How to Protect Data Beyond Passwords
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
How does the new model compare to the previous version?
The V4.1-Flash-1017 model reduces task execution time and output length while improving its DeepSWE v1.1 test score from 74.2% to 77%.
Are there costs associated with the new APIs?
The Search and Fetch APIs cost 75 cents per 1,000 successful requests, but calls through specific partner tools are free until November 1.
Did the release address recent outages?
The release notes focus on performance and agent tuning; DeepSeek separately promised to remove single points of dependency to prevent future outages.
