Why MiMo?
MiMo (Xiaomi) has become a serious contender. Xiaomi MiMo v2.5 delivers competitive reasoning at prices far below GPT-5. Through AI24X, access MiMo v2.5 Pro, MiMo v2.5 Flash alongside DeepSeek, Qwen, GPT-6 and more through one API key.
Pricing overview
MiMo v2.5 is priced in CNY and, unlike most tiers we track, splits input by whether the prompt was a cache hit. That one line separates a cheap MiMo bill from an expensive one: cached input costs a fraction of uncached input, and every agent workload re-sends the same system prompt and history on each turn.
| Line item | CNY / 1M | ≈ USD / 1M | When it applies |
|---|---|---|---|
| Input — cache hit | ¥0.02 | ≈$0.003 | Prompt prefix already cached |
| Input — cache miss | ¥1.00 | ≈$0.14 | New or changed prefix |
| Output | ¥2.00 | ≈$0.28 | Every generated token |
List prices as published by Xiaomi (verified 2026-08-22 against the official pricing page); CNY converted at ≈¥7.2/US$. The "from $0.14" figure used elsewhere on this site is the cache-miss input line. AI24X is an independent gateway and is not affiliated with Xiaomi.
Why the cache line dominates
Take one million input tokens. With no cache reuse that block costs ¥1.00. At a 90% hit rate it costs 0.9×¥0.02 + 0.1×¥1.00 = ¥0.118 — about 88% less for the same tokens. Output is unaffected at ¥2.00/M, so a generation-heavy job barely notices the change while a long-context agent job sees its input side collapse.
What a workload actually costs
| Workload | Input mix | Input | Output | Cost |
|---|---|---|---|---|
| Chat (1M in, 1M out) | 0% hit | ¥1.00 | ¥2.00 | ¥3.00 |
| Agent loop (1M in, 1M out) | 90% hit | ¥0.118 | ¥2.00 | ¥2.12 |
| Agent, long output (1M in, 3M out) | 90% hit | ¥0.118 | ¥6.00 | ¥6.12 |
| RAG, no reuse (4M in, 0.5M out) | 0% hit | ¥4.00 | ¥1.00 | ¥5.00 |
Two levers follow from the table: keep the stable part of the prompt at the front so the prefix stays cacheable, and stop re-sending history you can compress. Through AI24X, get managed credits at competitive rates or bring your own key; the pricing page carries the current published rates.
Best for coding, reasoning, multimodal tasks
MiMo (Xiaomi) excels at coding, reasoning, multimodal tasks. Teams using OpenAI SDK can switch by changing just the base URL.
Payment and signup from outside China
Getting direct access to Xiaomi MiMo from outside China often stalls on account and payment setup rather than the API itself. Through AI24X the path is shorter: create an account, top up AI Gateway credits with PayPal or an international card, and start calling — no Chinese phone number required. If you already hold a MiMo key you are happy with, the BYOK gateway keeps it in play behind the same OpenAI-compatible interface.
Migrating an existing codebase
Switching from an OpenAI-compatible setup is a two-line change: point base_url at https://api.ai24x.com/v1 and set the model id to MiMo v2.5 Pro or MiMo v2.5 Flash. Requests keep the same JSON shape, so streaming and tool-calling code that already works against OpenAI-style endpoints needs no rewrite. Run one smoke test before moving real traffic — the curl quickstart and OpenAI SDK guide walk through the first call and the settings that matter.
Common errors and fixes
- 401 Unauthorized — the key is wrong, truncated, or from another account. Re-paste the full key.
- 402 Payment Required — your AI Gateway credits balance has run out. Top up in the console and retry.
- 404 on the endpoint — the base URL must end at
/v1. Check the path segment is not missing. - Model not found — confirm the exact model id on the MiMo model page.
Getting started
- Sign up free
- Get an API key
- Set
base_urltohttps://api.ai24x.com/v1 - Start calling MiMo v2.5 Pro and other models
Compare with DeepSeek, other alternatives, and pricing comparison.
FAQ
Do I need a separate Xiaomi account?
No. One AI24X key covers MiMo v2.5 Pro, MiMo v2.5 Flash and the rest of the catalog, with billing and usage managed in one console.
Can I pay with PayPal?
Yes. AI24X accepts PayPal and international cards, and signup does not require a Chinese phone number.
Where do I check the current price?
Prices are configured in the platform and shown in the user console; the pricing page reflects the current published rates. Check there before budgeting a workload, since published rates can change.