Microsoft's new speech model is faster, cheaper than OpenAI's
What's the story
Microsoft has launched a new speech-recognition model, MAI-Transcribe-2, that is faster, more accurate, and cheaper than its competitors such as OpenAI and ElevenLabs. The tech giant has priced the service at just 10 cents per hour of audio. This is a massive reduction from the $0.36 per hour it charged for the first model in this line five months ago.
Cost savings
MAI-Transcribe-2 can lead to significant cost savings for enterprises
The new pricing model could translate into huge savings for enterprises.
For instance, a company processing 100,000 hours of call center audio annually would see its bill drop from $36,000 to $10,000 with the new model.
This makes transcription costs much more manageable and less likely to be contested within organizations.
Model enhancements
The model can transcribe audio in 60 languages
MAI-Transcribe-2 can transcribe audio in 60 languages, a significant improvement over the 43 languages supported by its predecessor.
It operates on Microsoft Foundry, the company's model marketplace for developers, and in MAI Playground, its testing environment.
The model is designed to handle real-world business audio with background noise, low-quality recordings, and overlapping speech rather than clean studio conditions.
User benefits
It also comes with advanced features like speaker diarization
MAI-Transcribe-2 also comes with some advanced features like speaker diarization, word-level timestamps, keyword biasing, and automatic language identification.
These capabilities are aimed at making the model more user-friendly for enterprises.
Speaker diarization helps identify who said what in a multi-person recording while word-level timestamps provide precise time markers for every word.
Benchmark results
The model has performed well on various benchmarks
MAI-Transcribe-2 has also performed well on various benchmarks. It ranks number one on FLEURS across 60 languages with an average word error rate of 5.2%.
The model is also second on the Artificial Analysis word-error-rate leaderboard and defines that firm's accuracy-latency Pareto frontier.
In terms of speed, it is 10 times faster than OpenAI's GPT-Transcribe, seven times faster than ElevenLabs' Scribe v2, and five times faster than Google's Gemini 3.5 Transcribe.
Strategic approach
Microsoft's aggressive strategy to make transcription a commodity
The launch of MAI-Transcribe-2 marks the third release in five months, each one expanding language coverage by about 40% while adding features that competitors usually charge extra for.
This rapid pace indicates a stable architecture and an aggressive strategy to make transcription a commodity before anyone else can.
Microsoft AI's CEO Mustafa Suleyman has credited this speed to a small, focused team that has been liberated from bureaucracy.