Whisper (OpenAI API)
Whisper (whisper-1) is OpenAI’s legacy hosted speech-to-text API, priced at $0.006 per audio minute. It is deprecated with a shutdown date of February 26, 2027, has a 25 MiB file ceiling, and does not support streaming. The open-source model remains available for self-hosting.
What is Whisper (OpenAI API)?
Whisper is OpenAI’s automatic speech recognition model, released as open-source code and weights. The hosted API surface for that model is the whisper-1 endpoint, which performs file-based transcription and translation. It is priced at $0.006 per minute of audio, rounded to the nearest second, with no separate free-minute allowance. The model supports multilingual speech recognition, speech translation, and language identification from a single file upload.
The critical context for anyone evaluating this tool today is its lifecycle status. OpenAI has marked whisper-1 as deprecated and announced a shutdown date of February 26, 2027. The deprecation covers the transcription endpoint and the translation endpoint, which only whisper-1 supports. This means the hosted API is not a viable long-term foundation for new work. The open-source model is a separate artifact: self-hosting the weights carries no license fee, but also carries compute, storage, and operational costs that the hosted API abstracts away.
Use Cases
- Transcribing pre-recorded interviews, meetings, or lectures where the audio can be chunked under 25 MiB and a one-pass transcript is sufficient.
- Translating speech from a supported source language into English text when using the /v1/audio/translations endpoint, understanding that the endpoint is deprecated alongside the model.
- Batch-processing an existing archive of audio files where the pipeline already integrates whisper-1 and the migration deadline has not yet been reached.
- Benchmarking or comparing the open-source Whisper model against newer transcription models before committing to a self-hosted deployment.
- Prototyping a speech-to-text feature on the open-source weights locally, then deciding whether to move to a supported hosted API or maintain a self-managed inference stack.
Pricing
Pricing model: Paid
| Plan | Price | Billing | Included usage | Notes |
|---|---|---|---|---|
| Pay-as-you-go (transcription) | $0.006 | per minute of audio | Transcription and translation via the whisper-1 model. Billed per minute of audio duration, rounded to the nearest second. No separate free-tier minute allowance is documented for this endpoint; usage draws against standard API tier rate limits (e.g., Free tier: 3 RPM / 200 RPD). | — |
Common Limitations
- Deprecation and shutdown · OpenAI has marked whisper-1 as deprecated with an announced shutdown date of 2027-02-26. The /v1/audio/translations endpoint supports only whisper-1 and is also scheduled for removal on the same date. New integrations should not be built around whisper-1; existing integrations need a migration plan before February 2027.
- File size ceiling · The whisper-1 transcription endpoint rejects requests larger than 25 MiB. Files above that limit must be split into chunks before transcription. This is a hard request-size limit, not a recommendation.
- No streaming support · whisper-1 does not support streaming. It processes a complete audio file and returns the full transcript; it cannot accept live microphone or call audio and emit text incrementally. For live transcription, OpenAI’s current documentation points to gpt-live-transcribe ($0.017/minute) or gpt-realtime-whisper, not whisper-1.
- Hallucination risk · OpenAI’s model card and independent research document that Whisper can produce text not actually spoken in the audio. One peer-reviewed study found roughly 1% of transcriptions contained entire hallucinated phrases or sentences, and 38% of those hallucinations included explicit harms. The model card cautions against use in high-risk domains where accuracy failures lead to pronounced outcome errors, and against transcribing recordings without consent.
- Not a real-time model · Whisper was not designed for out-of-the-box real-time transcription. Its speed and size make it suitable for file-based workflows; near-real-time use requires application-level engineering and is not a first-class capability of the hosted whisper-1 endpoint.
- Name confusion with third-party products · Multiple unrelated commercial products use “Whisper” in their branding (e.g., WhisperAI at whisperai.com, a separate speech-to-text SaaS with its own free plan and subscription pricing). The open-source Whisper model and the OpenAI whisper-1 API are distinct from those products. Pricing, limits, and data terms do not transfer between them.
Key Features
- File-based transcription of pre-recorded audio
- Speech translation into English via the translation endpoint
- Multilingual speech recognition and language identification
- Plain-text output for downstream processing
- Open-source model weights available for self-hosting
Whisper (whisper-1) vs. gpt-transcribe and WhisperAI
This comparison uses current official documentation and public pricing pages. whisper-1 is a deprecated hosted endpoint; gpt-transcribe is OpenAI’s recommended replacement for file transcription; WhisperAI is an independent commercial product that shares the “Whisper” name but operates on a different model, pricing, and feature set. The comparison focuses on documented lifecycle, limits, and cost structure rather than a subjective accuracy ranking.
| Tool | Positioning | Key differences | Best fit | Pricing note |
|---|---|---|---|---|
| Whisper (whisper-1, OpenAI API) Official whisper-1 model page Transcription deprecation notice OpenAI Audio API FAQ (25 MiB limit, no streaming) | OpenAI’s hosted speech-to-text API for the open-source Whisper model, in deprecation with a February 2027 shutdown. | Priced at $0.006 per audio minute; 25 MiB file ceiling; no streaming; plain-text output; translation endpoint available but also deprecated. The open-source weights are separate and unaffected by the API shutdown. | Legacy integrations already built on whisper-1 that are within their migration window, or teams benchmarking the open-source model before choosing a self-hosted or hosted path. | $0.006 per audio minute, billed to the nearest second. No dedicated free-minute allowance; usage is subject to standard API tier rate limits. Scheduled shutdown: 2027-02-26. |
| gpt-transcribe (OpenAI API) OpenAI transcription pricing Transcription workflow guide | OpenAI’s current recommended model for transcription of completed recordings. | Listed at $0.0045 per audio minute, lower than whisper-1’s $0.006 rate; positioned as the forward-looking replacement for file transcription; not subject to the same February 2027 shutdown as whisper-1. Output capabilities and migration details differ from whisper-1, so a direct model-ID swap does not preserve every feature. | New file-transcription integrations or whisper-1 migrations where the workload is completed audio and the team wants to stay on OpenAI’s supported transcription path. | $0.0045 per audio minute, as listed on OpenAI’s transcription price table. |
| WhisperAI (whisperai.com) WhisperAI about and feature overview | An independent AI speech-to-text platform for meetings, files, real-time dictation, and developer APIs, unrelated to OpenAI’s whisper-1 endpoint. | Offers a free plan with no credit card, unlimited transcription subscriptions starting at $24.99/month, and a developer API from $0.01 per minute[reference:12][reference:13]; supports files up to 5GB on eligible plans, speaker labels, AI summaries, translation, and meeting capture[reference:14][reference:15]. It is a separate product with its own infrastructure and terms, not an OpenAI API wrapper. | Teams that want a managed speech-to-text product with file and live transcription, speaker diarization, summaries, and a developer API under a subscription or per-minute model, rather than building on a deprecated OpenAI endpoint. | Free plan available; unlimited transcription starts at $24.99 per month; developer API pay-as-you-go from $0.01 per minute, with higher-volume subscription pricing. |
Comparison checked against official product information on 2026-11-08. Features and pricing may change.
Frequently Asked Questions
Is whisper-1 still available for new projects?
Technically the endpoint may still respond until the announced shutdown, but OpenAI has marked it deprecated and set a shutdown date of February 26, 2027. New integrations should not be architected around it. Existing integrations should budget migration work before that date, including application changes for any output features (such as timestamps or subtitle formats) that depended on the deprecated endpoint.
What replaces whisper-1 on the OpenAI API?
OpenAI’s current documentation points to gpt-transcribe for completed recordings ($0.0045 per audio minute) and gpt-live-transcribe for live audio ($0.017 per minute)[reference:7]. gpt-realtime-whisper is also listed as an option for existing integrations where supported, but it is not the same service as whisper-1 and is not recommended as the starting model for new integrations[reference:8]. A model ID swap does not preserve every output capability; the file-transcription compatibility documentation should be consulted for migration details.
Can I use whisper-1 for live transcription?
No. whisper-1 does not support streaming and processes a complete audio file. It cannot accept microphone or call audio as it arrives and emit text incrementally. For live transcription, OpenAI’s current guidance points to gpt-live-transcribe or realtime transcription endpoints, not whisper-1.
How accurate is Whisper, and what are the known failure modes?
OpenAI’s model card and independent research document that Whisper can hallucinate—produce text that was not spoken in the audio. One study found roughly 1% of transcriptions contained entire hallucinated phrases, with 38% of those hallucinations involving explicit harms[reference:9]. Hallucinations are more likely on lower-resource languages. OpenAI cautions against use in high-risk domains where accuracy failures lead to pronounced outcome errors, and against transcribing recordings without consent.
Is the open-source Whisper model affected by the API shutdown?
No. The deprecation applies to the hosted whisper-1 API endpoint, not to the open-source code and weights. Self-hosting the model carries no model-license fee, but the operator bears compute, storage, and operational costs. The open-source path is distinct from the hosted API in terms of pricing, maintenance burden, and data handling.
Alternative Tools
-
ElevenLabs ElevenLabs has the most expressive AI voice model in the market (Eleven v3) and the broadest language coverage among major TTS providers, but the Free plan is a non-commercial sandbox requiring attribution, Professional Voice Cloning is limited to your own voice even with consent, concurrency is a plan-level ceiling that credits cannot raise, v3 is explicitly not for real-time use, and Cartesia matches its naturalness at roughly half the price and a fraction of the latency. It is a quality-first purchase with rights, throughput, and legal constraints that determine whether it is usable for a given workflow. -
Suno Suno is an AI music generator with a free tier whose songs are permanently non-commercial and cannot be retroactively licensed, and paid tiers whose 20/60 monthly download caps are far tighter than their credit allowances—with full production workflows requiring Premier and Suno Studio. -
Udio Udio is a high-capacity AI music generator whose downloads have been disabled platform-wide since its October 2025 UMG settlement, whose terms grant no commercial-use rights on any plan, and whose output is owned by Udio or its licensors—making it a song-exploration lab rather than a production or distribution tool. -
Brave Search API Brave Search API provides web, news, image, video, local, and LLM Context results from Brave's independent index, priced per request with a $5 monthly credit, aimed at AI agents and RAG pipelines rather than Google SERP replication. -
Claude Code Claude Code is Anthropic's agentic coding tool that runs in the terminal, IDE, and desktop, powered by Sonnet and Opus models, with checkpointing, subagents, hooks, MCP support, and a native VS Code extension—available only through paid Claude subscriptions or API access, with no free tier. -
Grok Bot Grok Bot is xAI's AI teammate that runs tasks asynchronously in a cloud computer with browser, file system, and terminal access, available exclusively through The X Subscription's Premium tier and above.