Xiaomi is going all out! If you’re using OpenClaw or Hermes, pay attention to this great news: Xiaomi MiMo officially announced on May 27 that its MiMo-V2.5 series API will be permanently price-reduced starting today, with a maximum reduction of 99%, while also eliminating the long context multiplier pricing system, unifying billing for all input lengths. This is another massive price adjustment in the Chinese AI market, following DeepSeek’s permanent price cut of V4-Pro to 1/4 of its original price just three days ago. After the reduction, the overseas USD pricing for the MiMo-V2.5 series is nearly identical to DeepSeek’s, marking the two Chinese AI companies officially entering a head-on price battle on API pricing.
🚀 Better inference efficiency, lower costs, broader access.
MiMo-V2.5 Series API pricing is now permanently reduced — by up to 99% compared to previous pricing.
✨ Unified pricing across all context lengths.
MiMo Token Plans have also been upgraded:
• 5–8× more usable tokens… pic.twitter.com/E0C4MQLCzE— Xiaomi MiMo (@XiaomiMiMo) May 26, 2026
The MiMo-V2.5 series includes two versions: V2.5 and V2.5 Pro. The new reduced pricing no longer differentiates by context length—short prompts of 1K or long documents of 1M tokens both follow the same pricing formula, significantly simplifying the billing structure. At the same time, the Token Plan prepaid option has been upgraded, with available credits increased 5 to 8 times, and all consumed quotas fully reset. The voice synthesis API, MiMo-V2.5-TTS, remains free, offering an extra bonus for developers with voice needs.
Price War Dynamics: DeepSeek Ignites, Industry Follows
The direct trigger for this wave of price cuts came from DeepSeek. On May 24, DeepSeek announced that its flagship model V4-Pro would be permanently reduced to one-quarter of its original reference price, with the adjusted input price set at $0.435/MTok and output at $0.87/MTok. Just three days later, Xiaomi MiMo followed suit with its own price cut, matching DeepSeek’s USD pricing exactly.
Click here to sign up for the MiMo Token Plan (enter referral code PWR2ZZ to save $2)
Price reductions: up to 99% off
According to MiMo’s official announcement, calculated in RMB, the input cache hit price for V2.5 Pro has dropped to only ¥0.025 per million tokens, a reduction of 99%; the output price has dropped to ¥6.00/MTok, a reduction of 86%. The output price for V2.5 Basic has dropped to ¥2.00/MTok, a reduction of 93%.
In terms of global USD pricing, the MiMo-V2.5 series after its price cut nearly completely overlaps with DeepSeek’s pricing:
MiMo-V2.5 Pro (Overseas Pricing)
Input (cache hit): $0.0036 | Input (miss): $0.435 | Output: $0.87
MiMo-V2.5 (Overseas Pricing)
– Input (cache hit): $0.0028 | Input (cache miss): $0.14 | Output: $0.28
DeepSeek V4-Pro (after permanent price cut)
Input (Cache Hit): $0.0145 | Input (Cache Miss): $0.435 | Output: $0.87
DeepSeek V4 Flash
– Input (cache hit): $0.0028 | Input (cache miss): $0.14 | Output: $0.28
The numbers clearly show that MiMo-V2.5 Pro and DeepSeek V4-Pro have exactly the same input/output prices ($0.435/$0.87), and MiMo-V2.5 and DeepSeek V4 Flash are also identical ($0.14/$0.28). The only difference is in cache hit pricing: MiMo-V2.5 Pro’s $0.0036 is approximately 75% cheaper than DeepSeek V4-Pro’s $0.0145, which can save significantly more costs for scenarios with heavy cache usage (such as processing long documents repeatedly or multi-turn agent conversations). Additionally, MiMo-V2.5 offers Token Plan usage that DeepSeek doesn’t have, and the usage limits are directly increased by 5-8x without price increases—the reduction is quite noticeable.
Price Comparison: Chinese Models Now Form the Lowest Price Tier Globally
When looking at MiMo alongside DeepSeek, Chinese AI models have formed a highly competitive price tier in the overseas USD market. In contrast, pricing from major Western AI vendors remains several times higher.
OpenAI GPT-5.5
– Input: $5.00 | Output: $30.00 | Context: 128K
OpenAI GPT-5.4
– Input: $2.50 | Output: $15.00 | Context 128K
Claude Opus 4.7（Anthropic）
Input: $5.00 | Output: $25.00 | Context: 200K
Claude Sonnet 4.6（Anthropic）
Input: $3.00 | Output: $15.00 | Context 200K
Claude Haiku 4.5（Anthropic）
Input: $1.00 | Output: $5.00 | Context 200K
In the most direct comparison, MiMo-V2.5 Pro’s output price of $0.87/MTok is only 1/34th of GPT-5.5’s and 1/28th of Claude Opus 4.7’s. Even compared to Anthropic’s cheapest option, Haiku 4.5 ($5.00/MTok output), MiMo-V2.5 Pro is still nearly 5 times cheaper. In terms of context capacity, the MiMo-V2.5 series supports 1M tokens, which is 5 times Claude 200K and nearly 8 times GPT-5.5 128K.
The pricing of MiMo and DeepSeek is nearly identical, and this is no coincidence. Both companies have recently achieved significant improvements in reasoning efficiency, passing the cost savings on to developers. This marks a shift in China’s AI market from the past model of “burning money on subsidies” to “efficiency-driven” structural price cuts.
Token Plan Upgrade: Quota Increased 5-8x, Consumed Allocation Fully Reset
In addition to the real-time billing reduction for API, MiMo has also simultaneously upgraded its Token Plan prepaid option. Existing users’ available quotas have increased to 5–8 times the original amount, and all token quotas used within the validity period will be fully reset to zero for recalculation. This measure effectively gives developers who have already purchased annual or quarterly plans an additional budget subsidy.
The Tech Powering Price Cuts
In their announcement, MiMo stated that this price reduction is based on continuous inference efficiency optimization. The technical team fully supports SGLang HiCache-based SWA (Sliding Window Attention), reducing data transfer between GPU memory, CPU memory, and SSD multi-tier storage to 1/7 of the pre-optimization level, while increasing the number of cachable tokens to 5x the original amount. This significantly improves cache hit rates and inference efficiency (essentially leveraging DeepSeek’s open-source compression techniques). Additionally, through expert parallelism optimizations, the cluster’s input throughput capacity has been further enhanced.
Additionally, V2.5 has also been upgraded compared to V2-Pro in terms of model capabilities: improved long context retrieval quality (maintaining over 95% accuracy on needle-in-haystack tests in the 200K-800K range), reduced JSON error rate for parallel tool calling, and training data updated to Q1 2026. What’s even more attractive is that MiMo V2.5 Pro is actually considerably more powerful than DeepSeek V4 Pro, and it’s also a multimodal model that can recognize images and videos, making it even more compelling at the same price point.
For Taiwan developers, this price war brings tangible cost benefits. Take a typical Agent use case as an example: processing 1,000 conversations per day, with an average of 5,000 tokens input and 2,000 tokens output per conversation, the monthly cost using MiMo-V2.5 Pro is approximately $117 USD (about NT$3,750). The same workload on Claude Sonnet 4.6 costs approximately $1,350 USD (about NT$43,200), and on GPT-5.5 costs approximately $2,550 USD (about NT$81,600)—a difference of more than 11 to 22 times. This enables small and medium development teams that previously couldn’t afford top-tier models due to cost considerations to now easily access high-quality AI services. Moreover, with the Token Plan, if you’re only using it for Agent work on OpenClaw or Hermes, you won’t even be able to use it all up. The annual plan costs less than $64 USD, which comes down to just $5.33 USD per month for unlimited use—definitely worth considering.
Click here to sign up for the MiMo Token Plan (Enter referral code PWR2ZZ to save another $2)
Conclusion
Xiaomi’s MiMo-V2.5 price reduction marks the second major AI API price adjustment this week, following DeepSeek. After the cut, MiMo and DeepSeek’s overseas USD pricing are nearly identical, positioning Chinese AI models as the lowest-priced tier globally, with output prices at just 1/11 to 1/22 of Western major providers. For developers, options are expanding rapidly, and lower API prices mean cheaper AI applications and services. MiMo’s cache hit price ($0.0036/MTok) is more competitive than DeepSeek ($0.0145/MTok), making it potentially the more economical choice for agent applications relying on long-context caching.
Source: KOCPC Chinese