OpenAI's GPT Transcribe improves 0.7 points but trails ElevenLabs, Google, Mistral

OpenAI has introduced two new speech recognition models through its API: GPT Transcribe, designed for pre-recorded audio, and GPT Live Transcribe, built for real-time streaming. GPT Transcribe processes audio at approximately 34 times faster than real time, making it efficient for batch workloads. GPT Live Transcribe prioritizes low-latency real-time performance for live applications.

According to the AA-WER benchmark maintained by Artificial Analysis, GPT Transcribe achieves a word error rate of 3.31 percent. This represents a 0.7 percentage point improvement compared to GPT-4o Transcribe, its one-year-old predecessor. Concurrent with the model release, OpenAI reduced pricing by 25 percent, bringing the cost to $0.0045 per minute of audio. Both models accept text as transcription context, keywords, and support multiple input languages.

In the competitive landscape, OpenAI's transcription models rank fourth on the AA-WER benchmark. ElevenLabs Scribe v2 leads with a 2.3 percent error rate, followed by Google's Gemini 3 Pro at 2.9 percent and Mistral's Voxtral Small at 3 percent. Mistral has recently pressured the market by introducing Voxtral Transcribe V2 at $0.003 per minute, undercutting OpenAI's new pricing. These new transcription models complement OpenAI's recently announced Realtime model generation, which also includes the real-time transcription model GPT-Realtime-Whisper. Full technical details are available in OpenAI's Transcription Guide.

OpenAI's release of GPT Transcribe and GPT Live Transcribe represents a incremental but meaningful improvement in its transcription capabilities. The 0.7 percentage point reduction in word error rate and 25 percent price drop signal an effort to maintain competitiveness in a market where ElevenLabs, Google, and Mistral have established stronger benchmarks. ElevenLabs Scribe v2 leads the AA-WER ranking at 2.3 percent, while Google's Gemini 3 Pro (2.9%) and Mistral's Voxtral Small (3%) both outperform OpenAI's offering. More notably, Mistral's recent move to $0.003 per minute—underselling OpenAI's $0.0045 price point—appears to have prompted OpenAI's own pricing adjustment. The models' support for text context, keywords, and multiple input languages positions them as flexible tools, and their integration with OpenAI's recently announced Realtime model generation (including GPT-Realtime-Whisper) suggests the company is bundling transcription with broader real-time AI capabilities. Whether the combination of modest accuracy gains and aggressive pricing can close the gap with faster competitors remains to be seen.

No comments yet. Be the first to share your thoughts!