Alebex AI has ranked 8th in an international streaming speech-to-text accuracy benchmark from Artificial Analysis, published September 21, 2026 — ahead of speech models from Google, Microsoft, Nvidia and Amazon.
- Model
- Alebex STT (streaming speech-to-text)
- Word error rate
- 3.96% at final transcription
- Rank
- 8th worldwide of 29 streaming models on September 21, 2026; 1st in Canada, the only Canadian model listed
- Benchmark
- Artificial Analysis AA-WER Streaming Index, Final Transcription
- Snapshot
- September 21, 2026
- Built in
- Vancouver, British Columbia, Canada
Where Alebex sits on the leaderboard
A first for Canada
And we’re proud to be the first Canadian company represented in this category on the leaderboard.
What this is the start of
This is our first model release — and the first public piece of a much bigger effort to build real-time voice AI infrastructure in Canada.
For us, this is about more than a ranking. It’s about building world-class AI technology, expertise and intellectual property here in Canada.
Built in Vancouver.
Built in Canada.
This is only the beginning.
Frequently asked questions
What is Alebex STT?
Alebex STT is a streaming speech-to-text model built in Vancouver, British Columbia. It transcribes speech in real time over a single WebSocket, holding the whole turn in context rather than decoding fixed chunks.
How accurate is Alebex speech-to-text?
Alebex STT recorded a 3.96% word error rate for final transcription on the Artificial Analysis AA-WER Streaming Index, snapshot dated September 21, 2026. Word error rate is the percentage of words transcribed incorrectly, so lower is better.
Where does Alebex rank among speech-to-text models?
On the September 21, 2026 snapshot of the Artificial Analysis AA-WER Streaming Index, Alebex STT placed 8th worldwide out of 29 streaming models measured, ahead of speech models from Google, Microsoft, Nvidia and Amazon. Rankings change as models are added or retested.
Is there a Canadian speech-to-text model?
Yes. Alebex is the first Canadian company represented in the streaming speech-to-text category on the Artificial Analysis leaderboard. The model is built in Vancouver, British Columbia.
What is the AA-WER Streaming Index?
The AA-WER Streaming Index is an independent benchmark published by Artificial Analysis. It measures the percentage of words a streaming speech-to-text model transcribes incorrectly at final transcription, after end of speech has been detected.