Google: Chirp 3

Google: Chirp 3 is a speech-to-text model from Google, listed on api14 under the transcription category. It is the latest multilingual model in Google's Chirp line, and its provider description positions it as an upgrade in transcription accuracy over earlier versions. If your work involves turning recorded or streamed speech into written text, this is the family of model you would look at first.

The model takes audio as its only input and returns a transcription as its only output. According to the provider description, it covers 24 generally available languages and more than 77 preview languages, which makes it a genuinely multilingual option rather than an English-first system with a few extras bolted on. It also supports automatic language detection, so you do not have to declare the language of a clip before sending it, and automatic punctuation, so the returned text arrives with sentence boundaries rather than as one unbroken stream. No context length is published for this model, and the maximum output is listed as zero, which reflects the fact that transcription output is not measured the way chat completions are. Treat the audio you send as the real constraint rather than a token window.

The practical uses follow directly from those capabilities. Meeting and call transcription is the obvious one, especially where participants switch languages or where you cannot know the language in advance. Because language detection is automatic, you can point a single pipeline at mixed-language audio instead of routing clips by hand. Automatic punctuation makes the raw output far more usable downstream, whether you are feeding it into a summariser, a search index, a subtitle track, or a review tool for human editors. Voice note capture, interview transcription, and archive digitisation for multilingual material are all reasonable fits.

You call the model through api14's OpenAI-compatible API using the model id "google/chirp-3". Because the interface follows the OpenAI convention, existing clients and SDKs generally work with a change of base URL and model id rather than a rewrite. The model lists no supported request parameters, so plan on sending audio with the model id and relying on the model's own defaults for language handling and punctuation rather than tuning behaviour through extra fields. Check the live price table on this page for current rates before you commit to a volume.

One thing to weigh is scope. Chirp 3 does transcription and nothing else: it does not accept text input, it does not generate conversational replies, and it does not produce summaries or translations as separate outputs. If your task is chat, reasoning, or text generation, a general purpose language model is the right choice, and you would use Chirp 3 only for the audio-to-text step in front of it.