On May 27, 2026, Xiaomi cut the prices of its MiMo large-model API yet again — following DeepSeek's lead and pushing inference costs close to zero. But the part that matters most for builders on a budget is the part that didn't change: MiMo-V2.5-TTS is still free. If you've been waiting for a reason to try Xiaomi's speech model, this is it.
What just got cheaper
The cuts hit the MiMo-V2.5 line and take effect the same day. Cached input is now almost free, which heavily rewards prompt caching:
- MiMo-V2.5-Pro — cached input ¥0.025 / M tokens (down ~98–99%), uncached input ¥3 / M (down ~57–79%), output ¥6 / M (down ~71–86%).
- Standard MiMo-V2.5 — cached input ¥0.020 / M tokens (down ~96–98%), uncached input ¥1 / M (down ~64–82%), output ¥2 / M (down ~86–93%).
- MiMo-V2-Pro & MiMo-V2-Omni — API prices unchanged this round.
MiMo TTS is still free
Here's the headline for a free-API directory: MiMo-V2.5-TTS keeps its limited-time free-access policy. You get high-quality speech synthesis at zero cost — just bring an API key, no paid plan required. "Limited-time" with no announced end date, though, so the honest advice is to try it now rather than assume it'll stay free forever.
Why it matters
Two things stand out. First, the pricing structure makes prompt caching the difference between cheap and nearly-free — design your prompts to reuse a stable prefix and your bill collapses. Second, this is another data point in a polarizing market: Xiaomi and DeepSeek are racing toward near-zero inference pricing while the frontier labs hold a premium line. For anyone building side projects, that race is great news.
Try it
Grab a key and read the model docs on the Xiaomi MiMo platform. Xiaomi has also been running a free token-plan promotion through late May 2026, so it's a good moment to claim quota while it lasts.
And a small note from one MiMo fan: this is the first article in our directory, and we wanted to spend it on the team that keeps shipping real, genuinely-free models. Thanks, Xiaomi — keep going.