China’s DeepSeek has released a new version of its open-source artificial intelligence model, describing DeepSeek-V4.1-Flash as faster and leaner than its predecessor, with stronger benchmark performance and shorter outputs. The model was announced Saturday, with the company publishing its technical report on HuggingFace.
DeepSeek said the model uses reinforcement learning with self-feedback to improve end-to-end project work, from game creation to 3D modelling, and claimed substantial gains across reasoning, coding, agentic tasks and visual benchmarks. Because the model is open-source, developers can download and adapt it freely.
The launch comes with a caveat Bloomberg highlighted: DeepSeek recently suffered a five-hour outage that left its servers unreachable, and some experts attribute its growing slowdowns to insufficient computing resources as demand for its low-cost models surges.
The release sharpens the central contest in AI: China’s open-weight models, cheap and widely adopted by developers, against the proprietary systems of U.S. labs that lead on performance but cost far more to run. Each DeepSeek update tests whether that price gap can keep closing.
