Fact · Speech recognition
5.1percent word error rate (final transcript)
In Artificial Analysis's 2026 streaming benchmark, OpenAI's hosted GPT Realtime Whisper model improved from a 7.5 percent word error rate on the first partial transcript to 5.1 percent on the final transcript.
Checked against its source on October 17, 2026
In the source's own words
“led by Voxtral Mini Transcribe Realtime (9.6%→5.3%) and OpenAI GPT Realtime Whisper (7.5%→5.1%)”
Announcing AA-WER Streaming (Artificial Analysis), Artificial Analysis, Artificial Analysis, 2026, Key results, paragraph on which models improve by waiting for the final.
Open the original- What it measures
- GPT Realtime Whisper is OpenAI's hosted streaming model. It is a different product from the open-source Whisper models that you can run on your own computer.
- What it does not show
- Only a handful of the streaming models gained much from waiting for the final transcript. Most were already at their final accuracy by the first partial.
Source
Cited on
- AI Dictation: How AI Voice Typing WorksGuide
- Speech to Text and LLMs: What Language Models ChangeGuide
- Real-Time Speech Recognition: Streaming and LatencyGuide
- Can Whisper transcribe in real time?Question
- Can you transcribe a meeting live, in real time?Question
- How do you connect speech to text to an LLM?Question
- Which speech to text should you pair with an LLM?Question
- Why is AI dictation slower than built-in dictation?Question