IT Home On October 17, DeepSeek V4.1 Flash has been updated to the V4.1-Flash-1017 version. The official web interface, APP, and API model have been synchronized to the new model (the API call method and pricing remain unchanged; simply using deepseek-flash allows access to the latest model).
According to Deep Seek official statements, this update has achieved a Pareto improvement between performance and cost: in various benchmark tests, the capabilities of the new version of the model are significantly better than the previous V4.1 Flash version. At the same time, the task runtime and output length have been reduced, resulting in a decrease in the overall cost per call.
HLE (without tools): 42.6 (44.0)
Terminal-Bench 4.0: 42.3
DeepSWE v1.1: 77.0
ProgramBench: 29.1
NL2Repo-Bench: 68.2
Automation-Bench v1.0.6: 60.5
Agents' Last Exam: 35.4
Apex Bench v1.1: 72.1
Chartography (tools): 83.0
BabyVision (w/tools): 92.5
ZeroBench-main (w/tools): 51.0
DeepSeek said that the new model not only improves performance, but also reduces the task execution time and output length of the model, allowing users to complete tasks at a lower cost and faster speed, achieving a new performance-cost Pareto optimal under multiple Harness instances.
Additionally, DeepSeek has made specific adaptation optimizations for platforms such as DeepSeek Harness, Tencent CodeBuddy, and WorkBuddy. It is recommended that users upgrade DeepSeek Harness to version 0.2.1-rc.1 or higher to achieve full compatibility.
In terms of capability, V4.1-Flash-1017 introduces a notable change: through self-reflective reinforcement learning training, the model’s ability to build complete projects from scratch is enhanced, covering scenarios such as game development and 3D modeling.
The official statement says this improvement enables the model to handle more “wish-style” programming requests, that is, directly generating executable project results under extremely limited prompts and material references. Official demonstration cases show that in the DeepSeek Harness environment, the new version completed the entire project in very short interaction rounds.
At the same time, the model weights of DeepSeek V4.1-Flash-1017 have been made available for download. DeepSeek provides weight files on both HuggingFace and ModelScope, continuing the V4.1 Flash’s open-source approach under the MIT license.
IT Home Address:
HuggingFace: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017
ModelScope: https://modelscope.cn/models/deepseek-ai/DeepSeek-V4.1-Flash-1017