Whisper

Whisper is OpenAI's open-source speech recognition model, released in September 2022 and updated to Whisper v3 Large in late 2023. It is trained on 680,000 hours of multilingual web audio and achieves state-of-the-art transcription accuracy across 99 languages, with particularly strong performance on non-English speech compared to competing commercial services.

Whisper is a developer-oriented tool — it ships as a Python package with no consumer-facing application. Developers run it locally on their own hardware (GPU recommended for speed), integrate it via OpenAI's API, or use it through third-party wrappers that add user interfaces. Many of the transcription tools in this catalog (Descript, Otter.ai) use Whisper or Whisper-derived models as their underlying transcription engine.

Key capabilities include speech-to-text transcription in 99 languages, automatic language detection, translation to English from any supported language, timestamp-level output (word and segment), and multiple model size variants from tiny (fastest) to large-v3 (most accurate). The large-v3 model achieves near-human accuracy on clean audio in major languages.

Pricing: The model itself is free and open-source (MIT license) — downloadable from GitHub and runnable on local hardware. OpenAI's hosted Whisper API costs $0.006 per minute of audio, making it one of the most affordable commercial transcription APIs. There is no subscription — pay only for what you use.

Limitations: Whisper has no consumer-facing interface — users need either technical knowledge to run it locally or use a third-party wrapper. Local inference requires significant RAM and benefits substantially from a GPU. The API is usage-based with no usage dashboard built into a consumer product.

Best suited for developers building transcription into applications, researchers who need high-accuracy multilingual transcription, and technically capable users comfortable running Python scripts or using the API.