DeepSeek V4.1-Flash-1017 lowers AI task costs while outperforming rivals

AI model cuts per-task costs to about $0.10 while beating major competitors in coding benchmarks

DeepSeek released V4.1-Flash-1017 on October 17, 2026. This model achieves 72 percent accuracy in the DeepSWE benchmark at about $0.10 per task—significantly cheaper than its predecessor’s $0.50 per task for 74.2 percent accuracy. It outperforms Claude Opus 5, GPT 5.6-Sol, Kimi K3, and GLM 5.3 in coding and agentic tasks. The model is available via DeepSeek’s website, mobile app, and API, with open-weight versions on Hugging Face and ModelScope. It also provides Search and Fetch APIs for developers, with free access to these tools until October 31, 2026.

The model uses reinforcement learning with self-feedback (RLSF) training and is tuned for specific developer tools like Deepseek Harness and Tencent CodeBuddy. Its cost efficiency comes from optimized training that maintains high accuracy at lower per-task pricing.

This advancement moves access to agentic AI capabilities toward broader affordability. Developers can now run complex coding tasks at near-$0.10 per task—reducing barriers for small teams and individual users who previously faced high costs. The free API access period further supports early adoption.

Watch for Chinese government actions on open-weight model publishing, as reported policies could block releases. Independent benchmark validation remains absent, and benchmark results may not fully reflect real-world code quality. The model’s cost advantage is significant but unverified by third parties.

Source: The Decoder