Chinese AI startup DeepSeek has released a faster and lighter AI model, V4.1-Flash-1017, Bloomberg reported on Oct. 17.
DeepSeek said on social media platform X that V4.1-Flash-1017 is a small update to V4.1-Flash released in early September. It said benchmark performance is higher than the previous version, tasks finish faster and responses are shorter. The new model is available immediately on the web, the DeepSeek app and the developer platform.
DeepSeek said in a technical report posted on AI platform Hugging Face that performance improved significantly across reasoning, coding, agent tasks and visual benchmarks. It also said applications have broadened from game production to 3D modeling.
Bloomberg said competition is intensifying between Chinese open-weight models and U.S. closed models as DeepSeek releases new products in succession.