Microsoft launches MAI-Transcribe-2 speech model, claims superiority over Gemini 3.5 Transcribe and GPT-Transcribe at $0.10 per audio hour

Microsoft announced the debut of MAI-Transcribe-2, a speech recognition model aimed at developers needing transcription capabilities. The announcement positions the model as a direct alternative to existing services such as Gemini 3.5 Transcribe and GPT-Transcribe. No details about the model’s architecture or training data were included in the source.

The service is offered at a flat rate of $0.10 for each hour of audio processed, with the price guaranteed through the end of 2026. This pricing model allows engineers to predict operating costs for batch or streaming transcription workloads. By locking the rate for several years, Microsoft seeks to provide budget stability compared to variable-price alternatives. The flat-hourly rate is presented as a cost advantage over competing offerings.

Microsoft claims that MAI-Transcribe-2 outperforms Gemini 3.5 Transcribe and GPT-Transcribe in accuracy or speed, though the source does not specify the metric used. The claim rests solely on Microsoft’s internal evaluation, with no independent benchmark results shown. Engineers considering adoption would need to run their own validation tests to confirm the performance edge. Without external verification, the claim remains an unverified assertion.

Because the news appears in only one feed, there is no corroboration from other outlets or technical reports to substantiate the launch details. The lack of multiple sources increases the uncertainty around the model’s actual availability and feature set. Engineers should treat the announcement as a preliminary signal and seek additional documentation or trial access before committing to integration.