DeepSeek has announced a significant price reduction for their API services during the hours between half-past five in the evening to half-past seven the next day according to Thai time. The DeepSeek-V3 model has seen a 50% price drop, while the DeepSeek-R1 model has been slashed by 75%, making the prices of both models equivalent (though in reality, R1 has a “think” time that consumes more output).
The DeepSeek-R1 model has triumphed over OpenAI o1 in certain test scenarios, especially in mathematical and programming assessment categories. DeepSeek’s API costs are significantly lower than OpenAI’s, with o1 charging $15 per million tokens for input and $60 per million tokens for output. This price reduction will bring the cost down to less than 1% of OpenAI o1, but requires processing during late night hours, which may be suitable for batch processing scenarios. The input cost is around 5 dollars per million tokens, and with the added benefit of potential cache hits leading to even further reductions.
Source: DeepSeek
TLDR: DeepSeek slashes API prices by 50% for V3 model and 75% for R1 model during specific hours, offering cost-effective solutions compared to OpenAI o1. R1 model excels in tests against OpenAI o1, especially in math and programming. Lower input costs and potential cache hits make DeepSeek a more affordable choice for processing needs.
Leave a Comment