Microsoft MAI-Transcribe-2 hits near-top accuracy at $1.67 per 1,000 minutes
As seen on the 24/7 Wall St. homepage on September 3, 2026.
Microsoft is now selling near-best-in-class speech recognition for $1.67 per 1,000 minutes, squeezing the pricing power of standalone transcription vendors and reinforcing that it builds in-house rather than buying models.
Microsoft AI has released MAI-Transcribe-2, achieving 2.0% AA-WER (#2) at ~411x real time (#2) for $1.67 per 1,000 minutes, placing it on the accuracy-speed Pareto frontier at one of the lowest prices among high-accuracy models MAI-Transcribe-2 is @MicrosoftAI’s latest speech https://t.co/afCa90rbtI
- Replies1
- Reposts0
- Likes0
Continue ReadingShow less
MAI-Transcribe-2, Microsoft AI's latest speech recognition model, ranks second on accuracy among the models Artificial Analysis tracks, at a 2.0% AA-WER, and pairs that with throughput few competitors match.
At $1.67 per 1,000 minutes, MAI-Transcribe-2 is among the lowest-cost options in the high-accuracy tier. Artificial Analysis places it on the accuracy-speed Pareto frontier, so buyers do not trade precision against processing pace to get that price.
For standalone transcription vendors, a hyperscaler delivering near-best-in-class results at commodity pricing is a direct compression on their ability to charge a premium. Microsoft building this model in-house signals a continued push to own the full stack of AI capabilities within its own platform.
Investors in the transcription and speech-to-text space should note that this kind of pricing from a company with Microsoft's distribution reach tends to reset market expectations quickly, making it harder for pure-play competitors to justify higher rates on accuracy alone.
Mentioned: MSFT