AssemblyAI: Universal-3.5 Pro

AssemblyAI: Universal-3.5 Pro is a speech-to-text model from AssemblyAI, served through the provider's Sync API. It sits in the transcription category, so its job is narrow and specific: turning audio into text rather than handling chat, reasoning or generation tasks. If you are building something that needs a transcript, this is the kind of model you would evaluate alongside other transcription options.

The model takes audio as its only input and returns a transcription as its output. It produces a complete transcript with word-level timestamps in a single synchronous response, which means you get the timing information for each word without a separate alignment step. The Sync API path is designed for audio clips up to 120 seconds, so the model works on short recordings rather than long files. No context window is listed for this model, and the maximum output is recorded as zero, which reflects that it is not a text generation model with a token budget in the usual sense.

The practical uses follow directly from those traits. Word-level timestamps make it a fit for subtitle and caption generation, where each word needs a time offset. Short clips are common in voice notes, support call snippets, short interviews and voice memo transcription, so a synchronous call that returns the full transcript in one response suits interactive tools that cannot wait on a batch job. Because the response is synchronous, you can wire it into a request and response flow, for example a user uploads a clip and receives the transcript and timings back in the same interaction.

You call it through api14's OpenAI-compatible API using the model id "assemblyai/universal-3-5-pro". The request carries the audio input, and the response returns the transcript with word-level timestamps. The model lists no supported parameters, so you should not expect to pass sampling controls, temperature, top_p or similar generation settings; the transcription behaviour is fixed by the provider. Treat it as a transcription endpoint rather than a chat completion endpoint, and check the live price table on the page for current rates.

The main limit to keep in mind is clip length. The Sync API serves audio up to 120 seconds, so longer recordings fall outside what this model is set up to handle in a single synchronous call, and you would need a different approach or a different model for those. If your task is text generation, chat or anything beyond speech-to-text, this model is not the right fit, since its input is audio and its output is a transcription.