Xiaomi slashes MiMo-V2.5 API prices by 99%, joins DeepSeek's China AI price war

Xiaomi has permanently slashed prices on its MiMo-V2.5 series API by up to 99%, directly entering the price war recently escalated by DeepSeek. The new pricing eliminates context-window-based tiers—both 256K and 1M context lengths now share the same rate. For MiMo-V2.5, input cache hits cost just 0.02 yuan per million tokens (non-hit: 1 yuan), and output is 2 yuan. The Pro version is slightly higher: 0.025 yuan cache hit, 3 yuan non-hit input, and 6 yuan output. These figures closely mirror DeepSeek-V4-Pro's current promotional prices, which DeepSeek made permanent starting June 1st.

Beyond API cuts, Xiaomi overhauled its Token Plan subscription packages. Without changing subscription prices, the included Credits have been increased 5 to 8 times across all tiers. This move comes despite recent criticism of the industry price war by MiMo lead Luo Fuli. The adjustment puts Xiaomi's Lite plan in line with entries from Kimi, ByteDance, and Jieyue Xingchen, though Tencent's Hunyuan Hy still offers a 28 yuan monthly tier. In the high end, Xiaomi's Max plan is not the most expensive—Alibaba's premium is 1,398 yuan/month and ByteDance's Agent Plan Max is 950 yuan/month. At least five major domestic manufacturers (Xiaomi, DeepSeek, Alibaba, ByteDance, Zhipu, Tencent) have significantly altered their package systems in the past six months, with some reducing low-price options while increasing overall prices.