DeepSeek V4.1 Flash Gets New Model With Big Coding Gains, Cheaper Agent Tasks
Chinese AI company DeepSeek has released DeepSeek-V4.1-Flash-1017, an updated version of its Flash model with improvements in coding, agentic tasks, … Read More The post DeepSeek V4.1 Flash Gets New Model With Big Coding Gains, Cheaper Agent Tasks appeared first on ProPakistani .
DeepSeek, a Chinese AI company, has unveiled DeepSeek-V4.1-Flash-1017, an enhanced version of its Flash model boasting advancements in coding capabilities, agentic tasks, speed, and output efficiency. The updated model is accessible via DeepSeek's website, app, and API, with its weights made available on Hugging Face and ModelScope platforms.
DeepSeek asserts that this version outperforms its predecessor while delivering tasks more swiftly and generating shorter responses, presenting a Pareto improvement in the performance-to-cost ratio.
DeepSeek employs a post-training method called reinforcement learning with self-feedback (RLSF) to enhance the model's proficiency in end-to-end software development. This technique enables the model to utilize its own confidence scores or generated critiques as reward signals, rather than depending on human evaluators or external reward models.
Consequently, V4.1-Flash-1017 can construct complete projects from scratch, encompassing tasks such as game development, 3D modeling, design work, and even the creation of compact animated explainer videos.
The company fine-tuned V4.1-Flash-1017 for various development environments, including DeepSeek Harness (version v0.2.1-rc.1 or newer), Tencent CodeBuddy, and WorkBuddy. According to DeepSeek's own benchmarks, the new model surpasses its predecessor across all tested categories and outperforms larger systems like Claude Opus 5, GPT-5.6 Sol, Kimi K3, and GLM 5.3 in several coding and AI agent benchmarks.
At its lowest cost setting, V4.1-Flash-1017 achieved a top score of 77.0% on the DeepSWE benchmark, costing around $0.20 per task, while its predecessor scored 74.2% but cost approximately $0.50 per run.
The announcement coincides with China's potential tightening of controls on open-weight AI models and other critical AI technologies. The government's proposal might factor in whether a model ranks among the ten most capable systems when deciding whether additional restrictions apply. If implemented, this could pose challenges for enhanced reliance on Chinese open-weight models, potentially subjecting them to stricter export or publication regulations.
Written by urgent.news from ProPakistani's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
