Hangzhou-based startup DeepSeek released its V4.1-Flash-1017 model on October 17, 2026. This update follows the previous version released on September 10, 2026, marking a five-week development cycle. The company has shifted to a monthly update schedule throughout 2026, focusing on efficient architectures and open-source weights.
The new model achieved a 77.0% score on the DeepSWE v1.1 coding benchmark, an increase from the previous 74.2%. According to the company, this performance surpasses the 74.0% score of Anthropic's Claude Opus 5. The model is designed to produce shorter outputs, which reduces operational costs for users billed by text volume.
DeepSeek reports that its per-task costs are typically under $0.20–0.25, while competitor models range from $2 to $10+. The company utilized reinforcement learning with self-feedback to enhance the model's agentic and coding capabilities, allowing for the creation of games and 3D models.
For the first time, DeepSeek is offering external access to its Search & Fetch APIs, priced at $0.75 per 1,000 calls. Developers can access these tools for free until November 1, 2026. The model weights are available under an MIT license on platforms such as Hugging Face and ModelScope.
