OpenAI launches 2 transcription models at $0.0045 per minute

OpenAI's two new transcription models undercut its own whisper-1 on error rates while adding noise immunity, strengthening its grip on the developer API market.

"These models better understand context and can transcribe technical terms, numbers and accented speech more accurately," OpenAI Developers said in a post on X.

GPT-Transcribe costs $0.0045 per minute for recorded audio files, while GPT-Live-Transcribe costs $0.017 per minute for real-time processing. Both show lower word error rates than their predecessors — GPT-Transcribe beats whisper-1, and GPT-Live-Transcribe beats gpt-realtime-whisper, which launched in May 2026. OpenAI demonstrated the noise immunity by showing real-time transcription of speech while a musical instrument was being played. The models support multiple languages and can handle domain-specific terminology across industries including medicine, law and engineering.

The launch expands OpenAI's API revenue stream beyond text and image generation into voice transcription, a market where Google Cloud Speech-to-Text, Amazon Transcribe and Deepgram compete. OpenAI's pricing undercuts some rivals on an accuracy-adjusted basis, potentially pressuring pure-play transcription startups that lack the breadth of a full AI platform. Microsoft, OpenAI's largest investor and primary cloud partner, stands to benefit as Azure becomes the preferred deployment path for the new models.

The transcription market has been a battleground for AI companies since OpenAI's whisper-1 set a new accuracy benchmark in 2022. Google's Chirp model, part of Google Cloud AI, offers 119-language support with real-time streaming, while Amazon Transcribe Medical targets healthcare with HIPAA-compliant workflows. Deepgram's Nova-2 model, which claims 30 percent lower error rates than whisper-1 on certain benchmarks, charges $0.0043 per minute for batch transcription — nearly matching GPT-Transcribe on price.

OpenAI's move represents a broader push into voice AI, following its May 2026 launch of gpt-realtime-whisper for live captioning. The company has been expanding its API product surface area to capture more developer spend, competing directly with cloud hyperscalers that bundle transcription with broader cloud services. For investors, the key question is whether OpenAI can convert its AI brand advantage into sustained API revenue growth without triggering a price war that compresses margins across the sector.

GPT-Transcribe uses a time-based billing model, charging $0.0045 per minute of audio processed — or about $0.27 per hour. GPT-Live-Transcribe, optimized for low-latency streaming, costs nearly four times as much at $0.017 per minute, or about $1.02 per hour. The pricing differential reflects the higher compute cost of real-time inference versus batch processing.

The timing of the launch is strategic. Enterprise adoption of voice AI has accelerated as companies deploy AI-powered call centers, medical transcription and meeting summarization tools. OpenAI's models offer developers a single API for both batch and real-time transcription, reducing integration complexity compared to stitching together multiple vendor solutions. The company also benefits from its existing developer base — millions of developers already use OpenAI's API for text and image generation, creating a natural cross-sell opportunity.

For Microsoft, which has invested more than $13 billion in OpenAI and integrated its models into Azure OpenAI Service, the new transcription models add another revenue-generating capability to the cloud platform. Azure customers can access the models through the same API endpoints they already use for GPT-4 and DALL-E, lowering the barrier to adoption. Google and Amazon, which offer their own transcription services through Google Cloud and AWS, face increased competition from a vendor whose models have become the de facto standard for generative AI.

This article is for informational purposes only and does not constitute investment advice.