Whisper (OpenAI) Pricing 2026

OpenAI Whisper API costs $0.006 per minute ($0.36/hour) for both the legacy Whisper model and the newer GPT-4o Transcribe model. GPT-4o Mini Transcribe is available at $0.003/min ($0.18/hour) for cost-sensitive workloads. GPT-4o Transcribe with speaker diarization is also $0.006/min. New accounts receive $5 in free credits covering approximately 833 minutes of Whisper transcription.

OpenAI Whisper is available as both a free open-source model and a paid API. The open-source model can be self-hosted for free (but requires GPU infrastructure costing $276+/month). The API gives new accounts $5 in free credits covering approximately 833 minutes of transcription. After credits are exhausted, you pay $0.006/min for Whisper or $0.003/min for GPT-4o Mini Transcribe with no free monthly refresh.

OpenAI Whisper is a general-purpose speech recognition model trained on 680,000 hours of multilingual audio data. It is available both as an open-source model (for self-hosting) and as a hosted API through OpenAI's platform. Whisper supports 99+ languages for transcription and translation, handles various audio formats, and processes audio at 5-10x real-time speed. OpenAI has also released GPT-4o Transcribe and GPT-4o Mini Transcribe as successor models offering improved accuracy and features like built-in speaker diarization.

OpenAI Whisper costs $0.006/min vs Deepgram Nova-3 at $0.0043-$0.0077/min depending on plan. Deepgram is cheaper at scale with its Growth plan ($0.0043/min) and offers real-time streaming with sub-300ms latency, while Whisper processes at 5-10x real-time but is not designed for live streaming. Deepgram also provides $200 in free credits vs Whisper's $5. Choose Whisper for simple integration within the OpenAI ecosystem and batch transcription; choose Deepgram for real-time applications, lower per-minute costs at volume, and richer audio intelligence features.

OpenAI Whisper at $0.006/min is 75% cheaper than AWS Transcribe's base rate of $0.024/min. However, AWS Transcribe offers volume-based tiered discounts dropping to $0.0078/min at 5M+ minutes, narrowing the gap for very high-volume users. AWS also includes built-in features like speaker diarization, custom vocabularies, and PII redaction that Whisper lacks natively. Choose Whisper for straightforward, affordable transcription; choose AWS Transcribe if you need deep AWS ecosystem integration, custom vocabularies, or HIPAA-compliant medical transcription.

Yes, Whisper is open-source under the MIT license and can be self-hosted on your own infrastructure at no software cost. However, self-hosting requires GPU infrastructure costing approximately $276/month minimum for a dedicated GPU instance, plus DevOps overhead of $50-$200/month. The break-even point vs the API is roughly 500 hours of transcription per month. Self-hosting makes sense for organizations needing data sovereignty, HIPAA compliance, or processing extremely high volumes where the $0.006/min API rate exceeds fixed infrastructure costs.

The Whisper API supports mp3, mp4, mpeg, mpga, m4a, wav, and webm audio formats with a maximum file size of 25 MB per request. Files larger than 25 MB must be split into smaller chunks before upload. The API processes audio at 5-10x real-time speed, meaning a 60-minute file typically completes in 6-12 minutes. Billing is calculated per second of audio processed with no minimum charge per request.

Yes, OpenAI Whisper charges for the full duration of submitted audio including silence, music, and non-speech segments. A 60-minute file with 30 minutes of silence costs the same $0.36 as a 60-minute file of continuous speech. To reduce costs, preprocess audio with tools like FFmpeg to strip silence or use voice activity detection (VAD) before sending to the API. This can reduce costs by 20-40% for recordings with significant dead air, such as meeting recordings or surveillance audio.

Whisper / GPT-4o Transcribe and GPT-4o Mini Transcribe are usage-based API tiers with no base monthly subscription fee. You pay only for audio you process. Enterprise access via ChatGPT Enterprise or the API is custom-quoted. Multiple users report that actual credit consumption is difficult to predict and that OpenAI support cannot provide per-request usage breakdowns when disputes arise.

Based on user reports, hitting a usage cap blocks further API calls without advance warning. Users on paid OpenAI plans describe access being suspended mid-workflow for extended periods (one report: 18+ hours) with no compensation. If you need uninterrupted high-volume transcription, plan for potential interruptions or upgrade to the Whisper / GPT-4o Transcribe tier with appropriate rate limits.

User reports indicate that simultaneous multi-speaker audio is a known weak point. One user reported that two speakers talking at the same time caused the model to loop identical words repeatedly across multiple re-renders, consuming large credit volumes without producing accurate output. For multi-speaker diarization use cases, budget for higher-than-expected credit consumption and test thoroughly before production use.

In practice, refunds are rarely granted. Multiple users report that OpenAI refused refunds even when cancellation occurred within minutes of purchase and when EU consumer protection law (14-day cooling-off period) was cited. OpenAI support has not provided specific contractual or legal justification in these cases. Treat all API credit purchases and subscription charges as non-refundable.

Users consistently report that OpenAI support operates primarily through automated AI responses. Multiple reviewers describe receiving repetitive template replies, having cases closed without resolution, and being unable to reach a human agent — including on Pro-tier paid accounts. For billing disputes, users have resorted to chargebacks through their card providers.