Deepseek releases faster, cheaper AI model for software development tasks

Deepseek has released a new artificial intelligence model that performs coding and software development tasks more efficiently than its previous version. The company states that Deepseek-V4.1-Flash-1017 produces shorter outputs and completes work faster, which significantly lowers the cost per task for organisations using it for development projects.

According to The Decoder, the model is available through Deepseek’s platforms and as open weights on Hugging Face and ModelScope. Internal benchmarks show it outperforming larger rival models, including Claude Opus 5 and GPT 5.6-Sol, in several coding and agent categories. The company describes the update as a Pareto improvement in the balance between performance and cost.

What we know

The model was trained using reinforcement learning with self-feedback, a method that allows the AI to use its own confidence scores as rewards rather than relying on human evaluators. Deepseek says this approach improves end-to-end project development, including game programming and 3D modelling. The model is tuned for environments like Deepseek Harness, Tencent CodeBuddy, and WorkBuddy.

In the DeepSWE benchmark, V4.1-Flash-1017 achieved a top score of 77.0 per cent at an average cost of just under $0.20 per task. A lower setting cleared 72 per cent for about $0.10. The shorter outputs contribute to this efficiency, as noted in cost charts provided by the company.

Deepseek is also opening its internal search engine to developers via new Search and Fetch APIs. The standard price is $0.75 per 1,000 calls, but API users in tools like Claude Code and Codex receive free access for now. This marks the first time the company has exposed its in-house search engine to outside developers.

Defender's checklist

• Verify the source of downloaded model weights by checking official repositories like Hugging Face.
• Review API usage policies to understand pricing changes for search and fetch functions.
• Test the new model in isolated environments before integrating it into production workflows.
• Monitor output length and cost metrics to ensure efficiency gains match internal benchmarks.

The background in plain language

Teams usually evaluate AI models based on accuracy, speed, and cost when selecting tools for software development. Open-weight models allow organisations to inspect and modify the underlying code, which can improve transparency and security. Reinforcement learning with self-feedback helps models improve by critiquing their own work, reducing reliance on human annotation. Developers should check whether new models integrate smoothly with existing tools and whether cost savings justify any changes in workflow or compliance requirements.

What to do and how to stay safe: Deepseek

• Confirm that the model version matches the official release from Deepseek’s website or trusted platforms.
• Assess whether the new model’s output style aligns with your organisation’s coding standards and documentation needs.
• Monitor API costs closely, especially if you are using the free tier through third-party tools like Claude Code.
• Review security implications of using open-weight models, including potential vulnerabilities in downloaded weights.

Step-by-step guide: AI Governance for Small Teams: Practical Controls and Cost

General security guidance from the TechWire Beat newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

What is Deepseek-V4.1-Flash-1017?

It is an open-weight AI model released by Deepseek that improves coding and agent task performance while reducing costs.

How does it compare to other models?

According to Deepseek’s internal benchmarks, it beats larger rivals like Claude Opus 5 and GPT 5.6-Sol in coding and agent tasks.

Is there a cost to use the new APIs?

The Search and Fetch APIs cost $0.75 per 1,000 calls, but users of tools like Claude Code and Codex get free access for now.