Xiaomi MiMo for Real Estate Agents: One Family, Text and Voice

All Things AI · Season 3 · The New Models · Episode 6

MiMo is Xiaomi’s family of AI models, and it covers more ground than most: a huge flagship reasoning model, a cheaper fast version, a speed-tuned version, and a set of voice tools that read text aloud, design a brand-new voice from a description, or copy a voice from a short sample. You reach all of it through one API, or through Xiaomi’s own web chat and desktop app.

What it is

MiMo-V2.6 is Xiaomi’s current text model line, and it comes in three versions built for different jobs. Pro is the flagship: 1.02 trillion parameters total with 42 billion active on any given word, a million tokens of context, and it takes text, image, video, and audio as input. Flash is the same generation built for lower cost and higher call volume: 309 billion parameters total with 15 billion active, the same million-token context, the same four input types. UltraSpeed carries the same flagship Pro-level performance, tuned to answer up to twenty times faster, built for the moment you need the answer back in real time, not in a few seconds.

Alongside the text models sits a separate set of voice tools under the MiMo-V2.5-TTS name: reading text aloud in a set of built-in voices, designing a brand-new voice from a written description with no sample needed, and cloning a voice from a short audio sample. A speech-to-text model, MiMo-V2.5-ASR, rounds out the family.

How you get to it

Xiaomi’s own API platform is OpenAI-compatible, so any tool built for a standard chat API connection can point at it by changing one address. You create an account and generate a key there. The same models are also on OpenRouter under the names xiaomi/mimo-v2.6-pro, xiaomi/mimo-v2.6-flash, and xiaomi/mimo-v2.6-pro-ultraspeed, so if you already have an OpenRouter account, that is a second door to the same family.

Beyond the API, Xiaomi runs MiMo Studio, a web chat, and MiMo Desktop, a downloadable app, for anyone who wants a chat window rather than a connection to build.

The voice tools specifically

The text-to-speech, voice-design, and voice-clone tools sit on the same API platform as the text models, reachable through the same key. Reading text aloud needs a preset voice name. Designing a new voice just needs a written description of the voice in the request. Cloning a voice needs a short audio sample, sent along with the request, in MP3 or WAV.

Why one model handles text, image, video, and audio at once

Pro and Flash are both described as full-modality models: the same request can include text, an image, a video clip, or audio, and the model reasons across all of it together rather than treating each type separately. For an agent, that means a single request can hold a document, a photo, and a short clip together and get one coherent answer back, instead of running each input through a different tool and stitching the answers together yourself.

The million-token context window that both Pro and Flash carry means that combination can be large — a long document alongside several images — without running into the kind of context limit that forces you to break a job into pieces.

What to use it for

Pick the version by the job, not by habit.

  • Pro for the hardest, longest, highest-stakes drafting — a full document you need read closely and answered carefully.
  • Flash for the everyday volume: routine follow-ups, quick drafts, anything you’d otherwise ration because of cost.
  • UltraSpeed for anything where you’re waiting on the answer live and speed is the whole point.
  • The voice-design tool for a one-off voice you describe rather than record — useful when you want a distinct voice for something that isn’t Al or Victoria’s own.
  • The voice-clone tool where you already hold a short, consented sample of the voice you want to reproduce.

What it costs

On Xiaomi’s own published rates, Pro runs $0.435 per million input tokens and $0.87 per million output tokens, with a cached-token rate of $0.0036. Flash is cheaper at $0.14 in and $0.28 out per million tokens. UltraSpeed costs ten times the Pro rate — $4.35 in and $8.70 out — which is the price of the extra speed. The three text-to-speech models were listed as free for a limited time on Xiaomi’s own price page, with no end date published for that free period. The speech-to-text model, MiMo-V2.5-ASR, runs $0.074 per hour of audio.

Where this goes wrong

Two honest cautions.

Running everything through UltraSpeed by default. It costs ten times the Pro rate for speed you don’t need on a document you’re reading at your own pace. Save it for the moment you’re actually waiting live.

Assuming the free voice tools stay free indefinitely. Xiaomi’s own page calls it a limited-time rate with no stated end date, so check the current price before building a routine you depend on.

What to try in the first week

Two exercises, one for each half of the family.

Take a document you’d normally skim because reading it properly takes too long, and run it through Flash for a first pass. Notice what it costs, then decide whether Pro was ever actually necessary for that kind of job.

Separately, try the voice-design tool once: describe a voice in a sentence or two and listen to what comes back with no sample at all. It’s a fast way to understand what the tool is actually doing before you decide whether you’d ever reach for it.

The speech side, a little deeper

The three speech tools sit on their own API surface, separate from the text models but reachable with the same key. Reading text aloud in a preset voice needs nothing more than naming that voice in the request. Style itself can be steered by natural language or by an audio tag — asking for a specific emotional blend, or controlling tone at the paragraph level while controlling rhythm at the sentence level and stress at the word level. Output comes back as WAV or, for streaming, PCM16 audio at 24 kHz.

The speech-to-text model handles Chinese dialects plus mixed-language, ‘code-switch’ speech, which is a specific strength worth knowing about if you ever work with a multilingual recording.

How it sits next to the others

Against a single-purpose model, MiMo’s advantage is breadth in one place: a flagship, a fast tier, a speed tier, and a voice toolkit under one key instead of four separate accounts.

The tiered pricing — a flagship price, a fraction of that for the everyday tier, and a multiple of the flagship price for pure speed — is a pattern worth recognizing in any model family: match the tier to the job and the bill follows the job, not the other way around.

Questions agents actually type

What is Xiaomi MiMo?

A family of AI models from Xiaomi: MiMo-V2.6-Pro (the flagship, 1.02 trillion parameters), MiMo-V2.6-Flash (a cheaper, faster version), MiMo-V2.6-Pro-UltraSpeed (built for real-time speed), plus a set of text-to-speech, voice-design, and voice-clone tools, and a speech-to-text model.

How much does Xiaomi MiMo cost?

Pro runs $0.435 per million input tokens and $0.87 per million output tokens. Flash is $0.14 in and $0.28 out per million tokens. UltraSpeed is ten times the Pro rate. The text-to-speech tools were listed as free for a limited time, with no end date published.

How do I access Xiaomi MiMo?

Through Xiaomi’s own OpenAI-compatible API platform, where you create an account and generate a key, or through OpenRouter under the names xiaomi/mimo-v2.6-pro, xiaomi/mimo-v2.6-flash, and xiaomi/mimo-v2.6-pro-ultraspeed. There’s also a web chat, MiMo Studio, and a desktop app, MiMo Desktop.

Can Xiaomi MiMo design or clone a voice?

Yes. The voice-design tool builds a brand-new voice from a written description with no audio sample required. The voice-clone tool reproduces a voice from a short audio sample sent with the request. Both run through the same API as the text models.

Which MiMo model should I use for everyday drafting?

Flash. It’s built for high-frequency, everyday calls at a fraction of the flagship price, and for routine drafting from your own material it’s usually the right fit. Save Pro for the longest, hardest jobs.

Can MiMo handle images, video, and audio in the same request?

Yes. Both the Pro and Flash text models are full-modality — a single request can include text, an image, a video clip, or audio together, reasoned across as one input rather than run through separate tools.

What can Xiaomi’s speech-to-text model do?

MiMo-V2.5-ASR is officially open-sourced, with public weights and code, and it recognizes Chinese dialects plus mixed-language, code-switch speech. It’s priced at $0.074 per hour of audio on the API.

Book a 30-minute strategy call →

Season 1 is Claude Code for Real Estate Agents, filmed and on its own hub. All three seasons: AI for real estate agents.