DeepSeek AI: Cheaper Than OpenAI & Google

A new Chinese AI model is significantly undercutting the competition on price. Independent firm Artificial Analysis identifies this model as the most cost-effective serious AI system available. Developed by DeepSeek, the model completes a test suite for just three cents. This dramatically lowers the cost compared to models from OpenAI, Anthropic, and Google, while still performing competitively. DeepSeek charges as little as 14 cents per million input tokens and 28 cents per million output tokens for its V4-Flash model. This is considerably cheaper than most Western competitors. For repeated queries, input pricing drops by 98% to about three-tenths of a cent per million tokens. What's interesting is that DeepSeek's V4-Flash is 10 to 50 times cheaper than major American models like GPT-5.4 and Claude Opus 4.8 for similar performance. This is achieved through an efficient design, using a mixture-of-experts architecture. This design activates only a small portion of its parameters for each task, keeping computing costs low. The V4-Flash model even outscored DeepSeek's own larger flagship model on some coding and reasoning benchmarks. The bottom line: this aggressive pricing could force U.S. AI companies to reconsider their own pricing strategies.