Deepseek releases V4.1-Flash-1017, open-weight model beating larger rivals in coding tasks

Chinese artificial intelligence firm Deepseek has released Deepseek-V4.1-Flash-1017, an updated version of its Flash model. According to The Decoder, the model is available through the company's website, application, and application programming interface, with weights published openly on Hugging Face and ModelScope. This release marks a significant shift in performance metrics for agentic and software development tasks.

The new model utilizes reinforcement learning with self-feedback to enhance its capabilities. Deepseek states that this post-training method allows the system to use its own internal confidence scores as reward signals. This approach eliminates the need for human evaluators or external reward models, streamlining the training process for end-to-end project development ranging from game programming to 3D modeling.

Performance and Cost Efficiency

In internal benchmarks provided by Deepseek, V4.1-Flash-1017 outperforms its direct predecessor, V4.1-Flash. The model also surpasses larger competing systems, including Claude Opus 5, GPT 5.6-Sol, Kimi K3, and GLM 5.3, in several coding and AI agent categories. The company describes the update as a Pareto improvement in the tradeoff between performance and cost, offering better scores with faster task completion and shorter outputs.

Cost efficiency remains a central feature of this release. According to The Decoder, data from the DeepSWE benchmark shows the model achieving a top score of 77.0 percent at an average cost of just under $0.20 per task. In its low setting, the model clears a 72 percent score for approximately $0.10 per task. These figures highlight the impact of shorter outputs on overall operational expenses for developers.

Developer Tools and API Access

Deepseek is expanding its infrastructure by opening its in-house search engine to outside developers for the first time. The company has introduced new Search and Fetch APIs to facilitate this access. The standard pricing for these APIs is set at $0.75 per 1,000 calls. However, users of specific tools, including Claude Code and Codex, currently receive free access to these services, according to The Decoder.

The model is optimized for specific development environments. Deepseek recommends using Deepseek Harness version v0.2.1-rc.1 or later for optimal performance. The system is also tuned for integration with Tencent CodeBuddy and WorkBuddy. This focus on specific toolchains suggests a targeted approach to improving workflow efficiency for software engineers using these platforms.

What should you do?

Security and technology professionals should evaluate the open-weight release on Hugging Face and ModelScope for compatibility with existing development pipelines. Teams using Deepseek Harness should verify they are running version v0.2.1-rc.1 or later to ensure proper functionality. Organizations relying on external search capabilities should review the new API terms and pricing structures before integrating them into production environments.

What to do and how to stay safe: Deepseek

• Review the open-weight model documentation on Hugging Face to assess integration requirements for your current software development lifecycle.
• Verify that your Deepseek Harness environment is updated to version v0.2.1-rc.1 or later to support the new model features.
• Audit your current API usage patterns to determine if the new Search and Fetch APIs offer cost savings compared to existing solutions.
• Monitor the free access terms for Claude Code and Codex users to understand any potential changes or limitations in service availability.

Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation

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

Frequently asked questions

Is Deepseek-V4.1-Flash-1017 available as open source?

Yes, Deepseek has published the model weights as open source on Hugging Face and ModelScope, making it accessible to the broader developer community.

How does the new model compare to competitors in benchmarks?

According to Deepseek's internal benchmarks, the model outperforms Claude Opus 5, GPT 5.6-Sol, Kimi K3, and GLM 5.3 in coding and agent tasks.

What is the cost of using the new Search and Fetch APIs?

The standard cost is $0.75 per 1,000 calls, though users of tools like Claude Code and Codex currently have free access to these APIs.