DeepSeek-V4 Price Drops 90%, Millions of Tokens for 2 Cents - Liang Wenfeng Gives You Token Freedom
Last night, DeepSeek-V4 lowered its prices again, with the input cache hit prices for both models in the series directly reduced to 1/10 of the initial launch price. After the latest price adjustment, the input price (cache hit) for DeepSeek-V4-Flash is 0.02 yuan per million tokens, and DeepSeek-V4-Pro is 0.025 yuan.
On the evening of April 25th, DeepSeek-V4-Pro announced a 75% price reduction for DeepSeek-V4-Pro. Currently, the input price (cache miss) for DeepSeek-V4-Pro is 3 yuan per million tokens, and the output price is 6 yuan. This promotional activity will last until 23:59 on May 5th.
The other prices for DeepSeek-V4-Flash remain unchanged, with an input price (cache miss) of 1 yuan and an output price of 2 yuan per million tokens.
After the price reduction, DeepSeek has a greater price advantage compared to domestic large model companies:
▲ Comparison table of prices for domestic large model companies
Some Weibo users, based on their actual data, analyzed that this price reduction by DeepSeek equates to a 73% saving.
Another Weibo netizen exclaimed, "It feels like mobile data going from 5 yuan for 1GB to 5 mao for 1GB."
Some netizens speculated under DeepSeek's X account post that this is DeepSeek trying to gain market share because they have a technological advantage.
Some netizens bluntly stated that the token cache price war has begun, DeepSeek is too generous to us, and now is the best window for developers to evaluate migrating workflows from Claude or GPT to DeepSeek.
Other netizens affirmed the significance of DeepSeek's price reduction, believing that such a large discount will fundamentally change the economic landscape of production applications.
This time, DeepSeek-V4 has reduced the input cache hit price to 1/10 of the initial launch price, coupled with a 75% discount on the Pro version for a limited time, and with its advantages of being open-source and having a long context, it is rapidly attracting developers and application ecosystems, potentially meaning that small and medium-sized teams can also run business models with top-tier models.