September 22, 2026
Audio transcription got cheaper: Universal-3.5 Pro at 50% off
On September 22, OpenRouter made Universal-3.5 Pro available at 50% off through September 29. The model transcribes audio in 19 languages in a single synchronous request. It accepts recordings up to two minutes long and returns text, word-level timestamps, and a confidence score.
OpenRouter
@openrouter
Universal-3.5 Pro from @AssemblyAI is now available in OpenRouter: 50% off through September 29! It recognizes speech in 19 languages in a single synchronous request. According to independent speech recognition benchmarks, it ranks first in accuracy. Upload an audio file up to two minutes long and quickly get a complete transcript with word-level timestamps and a confidence score. Specify your domain's key terms and context in the prompt. Try it: http://openrouter.ai/assemblyai/universal-3-5-pro
· 14.3K views
Previously, OpenRouter accepted audio for transcription through POST /api/v1/audio/transcriptions and immediately returned text and usage without polling. Now Universal-3.5 Pro is available in the same call: you need a Bearer key, model, and input_audio. OpenRouter provides the exact request cost in usage.cost.
Vocabulary for the job. The model accepts WAV or raw PCM S16LE from 80 ms to 120 seconds. Audio URLs will not work: use base64 JSON or a multipart file up to 25 MB. Timestamps are enabled with timestamps=True, while a prompt and up to 1 000 keywords or phrases help the model handle domain-specific terminology.
The 50% discount runs through September 29.
