Xiaomi Slashes API Prices by 99% to Survive the DeepSeek Price War
Chinese tech giants are currently engaged in a brutal race to the bottom, practically giving away their AI models for pennies. It turns out that when your business model relies on copying others, the only way to compete is to destroy your own profit margins.
The corporate bloodbath started when DeepSeek made its massive promo discount on the flagship V4-Pro permanent, dragging down the industry standard for pricing. In response, Xiaomi decided to cut the pricing of its MiMo-V2.5 models by 99%, making it almost free. The company also completely removed any extra charges for long contexts, making the pricing flat across the board.
This aggressive move immediately pushed the MiMo-V2.5 model to the sixth spot on the global OpenRouter rankings. The sheer volume of traffic exploded, with the model processing 1.7 trillion tokens in just seven days, representing a massive 999% spike compared to the previous week.
The head of the MiMo team, Fuli Luo, who ironically used to be a key developer at DeepSeek, stated that their servers are running at near-maximum capacity. He claims the company is breaking even on these API sales solely due to clever cache optimizations.
Meanwhile, startup MiniMax decided to walk away from this suicidal price war. They released their flagship M3 model—which boasts open weights and scores 59.0 on the SWE-bench Pro benchmark, slightly outperforming GPT-5.5—but chose to introduce flexible subscription models instead of joining the race to zero.
This split has left cloud providers caught in the crossfire. Because their entire business relies on reselling API access, the sudden disappearance of margins means they are essentially hosting these massive models for free, while global developers are already using these ridiculous Chinese prices to squeeze discounts out of OpenAI and Anthropic.
Watching tech companies aggressively devalue their own products down to the cost of electricity is a glorious spectacle. While the executives pretend this is "cache optimization," the reality is a desperate, bloody scramble for market share where the first one to ask for actual money loses.
Source: South China Morning Post
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.