Whisper API
OpenAI's open-source speech-to-text
Overview
Whisper is OpenAI's open-source multilingual speech-to-text model — also available via the OpenAI API. State of the art for many languages when released. Widely used as a backbone for downstream transcription products.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 3 Sep 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
Use cases
What you can produce with Whisper API
- Transcribe audio files in dozens of languages to text via the OpenAI API or by running the open-source model locally.
- Translate non-English speech directly into English text in a single pass.
- Self-host the model with tools like whisper.cpp or faster-whisper to transcribe sensitive audio offline with no per-minute API cost.
- Generate timestamped output (SRT or VTT) to produce subtitles and captions for video content.
- Batch-transcribe large archives of recordings — interviews, lectures, podcasts — with a few lines of Python.
- Fine-tune the open-source model on domain-specific audio to improve accuracy for accents or specialist vocabulary.
- Build Whisper in as the speech-to-text layer of a larger app, such as a meeting-notes tool or voice interface.
Whisper is OpenAI's automatic speech-recognition model, available both as open-source weights and as a hosted API. It delivers excellent multilingual transcription and translation accuracy across noisy real-world audio, and the open weights mean teams can self-host for privacy or run it cheaply via the API at roughly $0.006/minute. It is a default choice for developers building transcription, captioning and voice pipelines.
It is a model, not a polished product: there is no diarisation, real-time streaming or speaker labelling out of the box, so you build the surrounding workflow yourself. Southeast Asian language coverage is decent for major tongues but weaker on low-resource regional languages. Best for technical teams comfortable with code; non-developers should look at a packaged transcription service instead.
What people say
Whisper remains one of the most consequential releases in speech technology: an open-source multilingual speech-to-text model that became the default backbone for a generation of transcription products. It is still available both as open weights (large-v3, plus the faster turbo variant) and through OpenAI's API. But its position shifted in March 2025 when OpenAI shipped gpt-4o-transcribe and gpt-4o-mini-transcribe as API successors, with benchmarks showing meaningfully lower word error rates than Whisper large-v3 at a similar price. OpenAI has not announced a Whisper deprecation, and the open-source model retains a huge advantage the newer API models cannot match: you can run it yourself, offline, for free.
Developer sentiment splits along that line. Whisper is praised for strong multilingual accuracy, permissive licensing, and a massive ecosystem — faster-whisper, whisper.cpp and countless wrappers make it easy to self-host on modest hardware, and for private or high-volume workloads it is still the pragmatic choice. The complaints are equally well documented: Whisper hallucinates. Researchers and engineers have reported fabricated sentences appearing during silences, with an AP investigation and the Koenecke et al. study documenting invented content — a serious problem for medical, legal or compliance transcription. It also lacks native speaker diarization and real-time streaming, forcing developers to bolt on extra tooling.
As a product entry, Whisper is best understood as infrastructure rather than an end-user app: it has no interface of its own, and most people encounter it inside other transcription tools.
Whisper genuinely fits developers who need self-hosted, multilingual transcription with no per-minute fees, and hobbyists transcribing personal audio locally. Teams wanting maximum accuracy via API should evaluate gpt-4o-transcribe instead, and anyone transcribing high-stakes audio must build in human review because of the hallucination record.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including Whisper API's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Whisper API unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Whisper API directly →
Spotted something out of date? Suggest an update →
More in Video & Audio
