Migrating from MiMo V2.5 to V2.6: Model IDs, Prices and API Examples

By og.cc.cd · Published 2026-10-20 · Updated 2026-10-20 · 5 min read

Xiaomi is retiring the MiMo V2.5 API models: after 10:00 on 21 October 2026 (Beijing time, UTC+8) the names mimo-v2.5-pro and mimo-v2.5 stop working. This guide maps the old IDs to the V2.6 series, lists the official prices, and shows the exact code changes.

Open the Cost Calculator →All guides

What is happening, and when

According to Xiaomi’s deprecation notice (Model Deprecation) and the email sent to platform users on 20 October 2026, there are two dates to know:

• System replacement — 14 Oct 2026, 18:00 (UTC+8). Since then, requests that still name mimo-v2.5-pro or mimo-v2.5 have been routed automatically to mimo-v2.6-pro and mimo-v2.6-flash, and billed at the new model’s price.
• Retirement — 21 Oct 2026, 10:00 (UTC+8). The old model names become invalid; requests that still use them receive an error.

In other words: your traffic is probably already running on V2.6. What breaks at the deadline is only the old model name — so the fix is a one-line change, but it has to ship before 10:00 Beijing time on 21 October (02:00 UTC).

Model ID mapping

The V2.6 series also includes mimo-v2.6-pro-ultraspeed, a faster mode of V2.6-Pro. The model list marks its rate limits as a custom service (contact Xiaomi), so most migrations will use Pro or Flash. Xiaomi asks that model names be written in lower case (V2.6 announcement).

Parameter changes to check

• Default max_completion_tokens: 32,768 on mimo-v2.5 → 131,072 on mimo-v2.6-flash. If you never set it, responses can now run much longer — set an explicit limit if you rely on short answers or a fixed budget. The value counts visible output and reasoning tokens; allowed range is 1–131,072 (Chat Completions reference).
• Thinking mode is enabled by default on the V2.6 models (thinking: {"type": "enabled"}; pass "disabled" to turn it off).
• temperature and top_p cannot be customised in thinking mode: the models force 1.0 and 0.95 even if you send other values.
• Multi-turn tool calls: in thinking mode the model returns reasoning_content alongside tool_calls; Xiaomi recommends keeping all previous reasoning_content in later requests for best results (First API call).

Prices (per 1M tokens)

Xiaomi says the V2.6 series keeps the V2.5 API prices (V2.6 announcement). The pricing page lists a mainland-China price in CNY and an overseas price in USD (Pay-as-you-go pricing). Real-time (standard) inference:

• USD = overseas price, CNY = mainland-China price, both as listed by Xiaomi (no currency conversion by us). No context-length tiers are listed.
• Cache hits are billed at the cache-hit price; cache writes are free for a limited time.
• Batch inference (Pro and Flash only) costs half: Pro $0.2175 / $0.435, Flash $0.07 / $0.14 (¥1.50 / ¥3.00 and ¥0.50 / ¥1.00). Ultraspeed does not support batch.
• Web search is billed separately per call, not in the token price.

Want a monthly number for your own traffic? The LLM cost calculator now includes all three V2.6 models (USD list prices), and the token counter shows per-prompt cost. MiMo token counts there are estimates based on o200k_base, not Xiaomi’s own tokenizer — check the usage field of real responses for exact numbers.

Context window and limits

All three V2.6 models list a 1M-token context window and a 128K maximum output. Pro and Flash share default limits of 100 requests per minute and 10M tokens per minute per account (summed across all your API keys for the same model). The text models support text generation, multimodal understanding, deep thinking, streaming, function calling, structured output and web search (Model list).

Endpoint and authentication

• OpenAI-compatible base URL: https://api.xiaomimimo.com/v1 (Chat Completions at /v1/chat/completions).
• Anthropic-compatible base URL: https://api.xiaomimimo.com/anthropic.
• Auth: either api-key: $MIMO_API_KEY or Authorization: Bearer $MIMO_API_KEY. Pay-as-you-go keys look like sk-…. Token Plan subscriptions use their own base URL and tp-… / ttp-… keys, and batch jobs use a dedicated batch base URL — copy those from the console (First API call).

Code examples

The only required change is the model string. Replace $MIMO_API_KEY with your own key (keep it in an environment variable, never in source control).

curl

curl https://api.xiaomimimo.com/v1/chat/completions \
  -H "api-key: $MIMO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mimo-v2.6-flash",
    "messages": [
      {"role": "user", "content": "Say hello in one sentence."}
    ],
    "max_completion_tokens": 1024
  }'
# Alternative auth header (also accepted): -H "Authorization: Bearer $MIMO_API_KEY"

Python (openai SDK)

# pip install -U openai
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["MIMO_API_KEY"],          # sk-... from the MiMo console
    base_url="https://api.xiaomimimo.com/v1",
)

resp = client.chat.completions.create(
    model="mimo-v2.6-pro",                        # was: mimo-v2.5-pro
    messages=[{"role": "user", "content": "Summarise this diff in 3 bullets: ..."}],
    max_completion_tokens=2048,                   # includes reasoning tokens
    # extra_body={"thinking": {"type": "disabled"}},  # thinking is enabled by default
)
print(resp.choices[0].message.content)
print(resp.usage)

Node.js (openai SDK)

// npm install openai
import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.MIMO_API_KEY,               // sk-... from the MiMo console
  baseURL: "https://api.xiaomimimo.com/v1",
});

const resp = await client.chat.completions.create({
  model: "mimo-v2.6-flash",                       // was: mimo-v2.5
  messages: [{ role: "user", content: "Classify this ticket: ..." }],
  max_completion_tokens: 1024,
});
console.log(resp.choices[0].message.content);

Migration checklist

1. Search your code, configs, environment variables and any agent/IDE settings for the old names (a grep one-liner is below the list).
2. Replace mimo-v2.5-pro → mimo-v2.6-pro and mimo-v2.5 → mimo-v2.6-flash (lower case).
3. If you were on mimo-v2.5, set max_completion_tokens explicitly — the default quadrupled to 131,072.
4. Decide whether you want thinking on (the default). If you need custom temperature/top_p, disable thinking — otherwise they are ignored.
5. Run your test suite and a few real prompts; compare output length and usage against what you saw before 14 October.
6. Check Billing → usage details in the MiMo console for any remaining calls to the old models.
7. Deploy before 21 Oct 2026, 10:00 (UTC+8) — that is 02:00 UTC, or 19:00 PDT on 20 October.

# find every hard-coded old model name before the cut-off
grep -rnE 'mimo-v2\.5(-pro)?\b' . --exclude-dir=node_modules --exclude-dir=.git

Nothing in this guide calls Xiaomi’s API from our site: the cost calculator and token counter run entirely in your browser.

Sources

Vendor features and prices were checked against these official pages on 2026-10-20. Products change; if something here no longer matches, the official page is right.

• Xiaomi MiMo — Model Deprecation (en-US)
  https://mimo.mi.com/docs/en-US/updates/deprecate
• Xiaomi MiMo — 模型下线 (zh-CN)
  https://mimo.mi.com/docs/zh-CN/updates/deprecate
• Xiaomi MiMo — Pay-as-you-go API pricing
  https://mimo.mi.com/docs/en-US/price/pay-as-you-go
• Xiaomi MiMo — 按量计费 API 定价
  https://mimo.mi.com/docs/zh-CN/price/pay-as-you-go
• Xiaomi MiMo — Model list
  https://mimo.mi.com/docs/en-US/quick-start/summary/model
• Xiaomi MiMo — First API call
  https://mimo.mi.com/docs/en-US/quick-start/summary/first-api-call
• Xiaomi MiMo — OpenAI Chat Completions API reference
  https://mimo.mi.com/docs/en-US/api/chat/openai-api
• Xiaomi MiMo — MiMo-V2.6 announcement
  https://mimo.mi.com/docs/en-US/news/latest/v2-6

Related guides

• How to Get an LLM API Key: OpenAI, Claude, Gemini, DeepSeek, Qwen, Kimi, GLM and More
• How to Get an OpenAI API Key: Steps, Region Rules, Billing and First Request
• How to Get a Claude API Key in the Claude Console (Anthropic)