DeepSeek launches V4-Flash with lower costs

Performance tests have shown that DeepSeek's Chinese V4-Flash model outperforms in terms of operating costs, with a cost of less than $0.14 per million tokens. This is significantly lower than the costs of models like GPT-5.6 and Claude Fable 5., which exceed tens of dollars. It achieves an approximate score of 50 on the AI intelligence index, reflecting high efficiency while substantially reducing expenses. The company relies on the "Mixture of Experts" technique, which allows only a portion of the model's parameters to be activated during use, thereby decreasing computational demands. The model has been trained on more than 32 trillion text tokens using advanced techniques to optimize performance and reduce costs. DeepSeek aims to regain momentum in the artificial intelligence market by offering highly efficient models at low prices, with plans to release updates and discounts to support developers.

Size: 120%Text Size

Notice: This Is an AI-Generated Summary

Display The Full Article