Xiaomi’s MiMo pricing has one feature that sets it apart from many cheap models: no long-context surcharge on the V2.5 series. Combined with a very large context window, that makes MiMo unusually affordable for the thing other models penalize — feeding a lot of code into a single request. This explains the current API rates, Token Plan angle, free access windows, and how to estimate cost.
For setup, see run Xiaomi MiMo V2.5 Pro with Claude Code.
Current API rates
Xiaomi’s official V2.5 price update, effective May 27, 2026 at 00:00 (GMT+8), lists these pay-as-you-go rates:
MiMo V2.5 API pricing (per 1M tokens, checked June 9, 2026)
| MiMo-V2.5 | Input cache hit: $0.0028; input cache miss: $0.14; output: $0.28 |
|---|---|
| MiMo-V2.5-Pro | Input cache hit: $0.0036; input cache miss: $0.435; output: $0.87 |
| MiMo-V2.5-TTS series | Listed as free for a limited time |
| Legacy MiMo-V2-Pro / Omni | Auto-routes to V2.5 pricing from June 1, 2026; deprecated June 30, 2026 |
Why “no surcharge” matters
Most models that advertise a big context window quietly charge more once your prompt crosses a threshold — sometimes multiplying the rate. For agentic coding, where you want to load lots of files, that penalty adds up fast.
MiMo dropping the surcharge means the ~1M token context is genuinely usable on a budget. You can feed large parts of a repo and pay the same per-token rate rather than a higher long-context tier, which is exactly the workflow Claude Code and Codex benefit from on big projects.
The model tiers
MiMo tiers (verify on Xiaomi)
| MiMo-V2.5-Pro | Flagship reasoning model; ~1M context; $0.435 cache-miss input / $0.87 output per 1M |
|---|---|
| MiMo-V2.5 | Standard V2.5 model; ~1M context; $0.14 cache-miss input / $0.28 output per 1M |
| MiMo-V2.5-TTS series | Speech models; listed as free for a limited time |
| Legacy MiMo-V2 names | Auto-route to V2.5 during the June 2026 migration window |
For agentic coding with big context, V2.5 Pro is the stronger pick. For cheaper everyday agent turns, start with standard MiMo-V2.5 and escalate to Pro when reasoning quality matters.
Free trial access
MiMo has launched with free trial access through various coding tools — OpenCode, Cline, KiloCode, and others at different windows. These are promotional and come and go, so the way to catch them is to check Xiaomi’s platform and the integrations in your tool of choice. If a free window is open, it’s a no-cost way to evaluate MiMo before paying.
Estimating your cost
Estimate with cache hits separated from new input:
- Estimate cache-hit input tokens.
- Estimate cache-miss input tokens.
- Estimate output tokens.
- For MiMo-V2.5: hit input × $0.0028/M + miss input × $0.14/M + output × $0.28/M.
- For MiMo-V2.5-Pro: hit input × $0.0036/M + miss input × $0.435/M + output × $0.87/M.
Cache hits are dramatically cheaper, so repeated agent sessions that reuse the same context can cost far less than fresh prompts. Output is still the expensive side, so concise, targeted requests keep costs down.
Before you budget MiMo
- Confirm the current V2.5 and V2.5-Pro rates on Xiaomi's platform
- Note there's no long-context surcharge
- Check for current free trial windows in your tool
- Pick V2.5 Pro for harder big-context coding; test V2.5 for cheaper everyday use
- Estimate cache-hit input, cache-miss input, and output separately
Wrapping up
Xiaomi MiMo’s current V2.5 pricing is built around very low cache-aware token rates with no long-context penalty, which makes its ~1M token context genuinely affordable for big-repo coding. Free trial windows through coding tools let you test it first. Start with MiMo-V2.5 for cheap runs, move to MiMo-V2.5-Pro for harder agentic work, and confirm the live rate on Xiaomi’s platform.
To put it to work, see run Xiaomi MiMo V2.5 Pro with Claude Code. To compare cost models, see coding plans vs pay-per-token.
