Questions? hello@viberation.devGet supportBlogDocsChangelog
Get started

MiMo

Xiaomi's open-weight family, and unusually multimodal for its price.

MiMo is Xiaomi's model family, published as open weights and built around a flagship at over a trillion parameters with Flash variants under it. The whole line reads text, images, video and audio with a million-token context, which is a rarer combination than the price suggests. Worth a look if your app takes messy real-world input — screenshots, recordings, clips — rather than clean text.

At a glance — MiMo-V2.5

Cost
$0.14 in · $0.28 out per 1M tokensBudget
Memory
1.1M tokens — about 1,600 pages
Reads
Text · Images · Audio · Video
Can do
  • Use tools (agents)
  • Structured output
  • Thinking mode
Good at
Coding: better than 65% · Agentic work: better than 46%
Reliability
99.86% uptime, last 24h
Released
Apr 22, 2026
Advanced details
API model ID
xiaomi/mimo-v2.5
Max output
131K tokens
Cached input
$0.003 per 1M
Tokenizer
Other
Coding index
56.8
Agentic work index
15.8

Providers

The same model, hosted by different companies. OpenRouter picks one per request and falls back to the next if it fails.

ProviderInput /1MOutput /1MCached /1MMax outputUptime 24h
GMICloud · fp8Degraded$0.119$0.238$0.003945K85.61%
DeepInfra · fp8$0.133$0.266$0.003131K99.86%
Xiaomi · fp8$0.14$0.28$0.003131K98.03%
StreamLakeDegraded$0.168$0.336$0.003128K81.16%
Novita · fp8Degraded$0.168$0.336$0.003131K93.62%
Venice · fp8$0.40$2.00$0.0866K88.79%

Supported parameters

  • frequency_penalty
  • include_reasoning
  • logit_bias
  • max_tokens
  • min_p
  • presence_penalty
  • reasoning
  • repetition_penalty
  • response_format
  • seed
  • stop
  • structured_outputs
  • temperature
  • tool_choice
  • tools
  • top_k
  • top_p

Live from OpenRouter, refreshed hourly. Scores from Artificial Analysis.

Key info

Pricing
Open source
Category
Models
Best for
Expert
Made by
Xiaomi
Weights
Open
Good for
Apps that take images, video or audio

More in Models

  • France's frontier lab, with the small models everyone self-hosts.

    Models
  • Meta's current line, built for long multi-agent runs.

    Models
  • The open-weight family everyone else gets benchmarked against.

    Models
  • Google's open models — Gemini's research, small enough to self-host.

    Models