Notice that deepseek-v4.1-flash was upgraded on October 27, 2026 and now points to the deepseek-v4.1-flash-1017 snapshot, with billing and pricing unchanged.
DeepSeek-V4.1-Flash update notice
2026-10-27 (depending on the actual release time)
Effective October 27, 2026, QwenCloud will upgrade the deepseek-v4.1-flash model. After the upgrade, the deepseek-v4.1-flash endpoint will automatically transition to the snapshot version deepseek-v4.1-flash-1017, with billing items and pricing remaining unchanged. DeepSeek-V4.1-Flash-1017 is the latest iteration of V4.1-Flash. Trained with self-feedback reinforcement learning (RL), it significantly improves on benchmarks including Terminal-Bench 4.0, DeepSWE v1.1, and ProgramBench, while reducing task runtime and output length, reaching a new performance-cost Pareto optimum under multiple harnesses. Its ability to build complete projects from scratch is further enhanced, supporting wish-driven coding tasks such as game development and 3D modeling. It retains the previous version's 552B total-parameter MoE architecture, native multimodal visual understanding, 1M context window, and 384K maximum output.
You can test the model performance by specifying the model name as deepseek-v4.1-flash-1017 before October 27, 2026. Log on to QwenCloud to try it out. For more model details, see deepseek-v4.1-flash and deepseek-v4.1-flash-1017.
On this page