Tasks and inputs
MAI Realtime 1 is listed for Audio-to-audio, Text → Speech, Speech → Text. It accepts audio, text and produces audio, text. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 18, 2026
MAI Realtime 1 is Microsoft AI's first speech-to-speech model for live voice agents, from $0.01 per audio minute, in limited preview via MAI Labs and Foundry.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Indicative: from $0.01 per audio minute for speech-to-speech; reasoning $1 input / $5 output per 1M tokens. No billing during limited preview. Confirm at onboarding.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Voice agents fail in small, human ways. They talk over people, miss a quick "mm-hm", or answer before the caller has finished. MAI Realtime 1, Microsoft AI's first speech-to-speech model, announced on 17 November 2026, is built around those moments: it handles interruptions, reacts to short acknowledgements, and is designed to wait when the speaker has not finished.
It is not broadly available yet. Access runs through an interest form for MAI Labs or a limited-access preview in Microsoft Foundry.
Pricing has two meters, which is unusual enough to plan for:
The split exists because of how the model works. A fast system handles real-time turns and escalates to a stronger reasoning engine when a request gets hard. Calls that stay simple should stay near the audio rate; calls that need thinking add token charges. Preview customers are not billed until public preview begins, and all of these figures are indicative.
Microsoft built it to be compatible with the OpenAI Realtime API. That lowers the cost of a trial considerably if you already run GPT-Realtime-2.1: point an existing client at the new endpoint and compare. Microsoft claims MAI Realtime 1 beats GPT-Realtime-2.1 on quality, speed and price, and sits on the price-performance frontier for Big Bench Audio and Tau-Voice. No scores have been published, so that claim is untestable from the outside for now.
Languages, the voice list and latency figures have not been published. Neither has a general availability date. If your use case depends on a specific language or a hard latency target, ask during onboarding before you build.
Record twenty real calls from your current system, with permission, and script them as test conversations. Include at least five with barge-in, a noisy line and one that requires a lookup in a backend system. Score turn-taking errors by ear, not just task success. Then check how many calls triggered the reasoning meter, because that ratio decides your real cost per minute. Microsoft's existing voice output model, MAI-Voice-2.1, and the wider audio-to-audio category are useful reference points.
Compare the model on the work you need to complete.
MAI Realtime 1 is listed for Audio-to-audio, Text → Speech, Speech → Text. It accepts audio, text and produces audio, text. Use the publisher’s input format and preprocessing for your chosen deployment.
Compare pronunciation, voice consistency and time to the first playable audio. Check the service’s billing unit for the actual script.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
OpenAI
GPT-Live-1.1 is OpenAI's full-duplex Live API voice model, with more accurate transcription and steadier multilingual accents than GPT-Live-1.
rime
Rime Prism is a speech-to-speech model for enterprise voice agents that settles each reply as inspectable text first, with parallel tool calls.
Liquid AI
LFM3-Audio-2B is Liquid AI's open 2B speech model: ASR, TTS, voice chat and interleaved text-and-speech output from one local checkpoint.
Alibaba
Qwen-Audio-3.1-Realtime Flash is Alibaba's lighter full-duplex speech model with audio and text in and out, function calling, web search and a 262K context.
ElevenLabs
Eleven Realtime v1 Max is ElevenLabs' reasoning speech-to-speech agent for support and multi-step tool workflows, OpenAI Realtime-compatible.
ElevenLabs
Eleven Realtime v1 Mini is ElevenLabs' fast, low-cost speech-to-speech voice agent stack, with an OpenAI Realtime-compatible API and 90+ languages.