You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.

AssemblyAI's Universal-2 is a speech recognition model: you upload an audio recording, and you get text as output. According to our data, it supports working with large audio files, so the model is suitable for transcribing interviews, calls, podcasts, lectures, and voice messages.

In a single response, the model returns up to 4096 tokens, with a context of 4095 tokens. This may not be enough for a very long recording, so it is more convenient to transcribe such files in parts: break the audio into fragments, and then assemble the finished text.

In our data, step-by-step reasoning is not noted for Universal-2; the model's task is different — to convert audio to text. If the transcript needs to be analyzed, summarized, or structured, switch to a text model in the same BotHub window and pass the transcript to it.

In our data, function calling and structured JSON output are not noted for Universal-2. For scenarios requiring strict response schemas or access to your tools, use text models via the unified OpenAI-compatible BotHub API: you won't have to rewrite the integration when switching models.

Audio is the main format for this model: you send a recording, you get text, including for large files. The acceptance of images and video is not noted in our data, so for such materials, choose specialized models in the BotHub catalog; switching takes a couple of seconds.

We do not have data on supported languages, so we cannot promise accuracy in Russian — it is more reliable to check with your own recording. You pay for the caps used, tokens do not expire, and access from Russia works without a VPN or a foreign card.