Price Reduction Takes Effect
Xiaomi's MiMo-V2.5 series APIs have been permanently reduced in price starting today. The new pricing took effect at 00:00 Beijing Time on May 27, with prices denominated in ¥/million tokens.
Both models received significant cuts. MiMo-V2.5-Pro output pricing dropped from ¥21-42/M to ¥6/M, while MiMo-V2.5 output fell from ¥14-28/M to ¥2/M. The largest reductions came in cache-hit scenarios: the Pro model saw 98-99% drops, and the standard version saw 96-98% drops.
MiMo-V2.5-Pro Pricing Details
| Item | New Price (¥/M) | New Price ($/M) | vs <256k Reduction | vs 256k-1M Reduction |
|---|---|---|---|---|
| Input (Cache Hit) | 0.025 | 0.004 | ↓98% | ↓99% |
| Input (Cache Miss) | 3.000 | 0.441 | ↓57% | ↓79% |
| Output | 6.000 | 0.882 | ↓71% | ↓86% |
MiMo-V2.5-Pro is Xiaomi's flagship reasoning model, with HuggingFace weights around 963GB. It has performed close to Claude Opus 4.6 in several third-party comprehensive benchmarks. At this price point, the input cost is already lower than GPT-5.5's $2.50/M.
MiMo-V2.5 Pricing Details
| Item | New Price (¥/M) | New Price ($/M) | vs <256k Reduction | vs 256k-1M Reduction |
|---|---|---|---|---|
| Input (Cache Hit) | 0.020 | 0.003 | ↓96% | ↓98% |
| Input (Cache Miss) | 1.000 | 0.147 | ↓64% | ↓82% |
| Output | 2.000 | 0.294 | ↓86% | ↓93% |
MiMo-V2.5 weighs in at around 295GB and is positioned as a lightweight reasoning model. At ¥2/M for output, it's among the lowest-priced large model APIs in China.
(USD prices converted at $1 = ¥6.80)
Other Products
The MiMo-V2.5-TTS series speech synthesis models continue to be available for free during a limited-time promotion. MiMo-V2-Pro and MiMo-V2-Omni, the two older models, maintain their original API pricing.
What This Means
The first half of 2026 has seen fierce API pricing competition among Chinese large model providers. DeepSeek V4, GLM-5.1, and Kimi K2.6 launched in quick succession, each competing for developer mindshare. Xiaomi's pricing cuts are quite aggressive, especially for cache-hit scenarios — prices of ¥0.025/M and ¥0.020/M are practically free.
For developers whose applications involve high-frequency calls (agent loops, code completion, real-time conversation), the MiMo-V2.5 series now offers excellent cost efficiency. Cache-hit pricing means repeated calls cost virtually nothing.
Xiaomi has been investing in large model R&D since 2023, taking a full-stack approach. Getting costs down to this level shows they've done serious work on inference optimization.