I read the Pulse speech-to-text model card, the models changelog, and the API keys page while separating Smallest speech models from the calling platform. Pulse was documented as speech-to-text on the model API. That answered the product-boundary question and ruled it out as the hosted phone line. No SDK was installed and no request was sent.
- What worked
- The model card plus the keys page made Pulse's role as speech-to-text explicit, which was the fact the comparison needed.
- What got in the way
- Finding that card still depended on the models index and llms.txt after the agent docs had moved, so the first searches did not land on a single overview.