Retell sits in a crowded category of voice agent platforms and it competes on the two things that actually matter for phone calls, latency and telephony integration. The latency claim of roughly 600ms is credible in my testing and it matters more than people expect. Below about 800ms a caller experiences a conversation. Above it they experience waiting, and they start talking over the agent, and turn detection falls apart, and the whole interaction degrades.
Getting this right is unglamorous engineering and Retell has done it. The telephony side is the other reason to pick this over rolling your own. SIP trunking support means you can connect an existing phone system rather than handing out a new number, which is the difference between a pilot somebody in operations will actually run and a demo that stays in the demo folder. Warm transfer to a human works, live call monitoring works, and post-call transcripts with custom analytics are there without you building a pipeline.
Now the part I want to be direct about, because it is where people get caught. The advertised range of $0.07 to $0.31 per minute is not a tier structure, it is the spread between a cheap configuration and an expensive one, and the total is assembled from voice infrastructure at about $0.055, text to speech at about $0.015, telephony at about $0.015, and then the language model, which is where the range lives. A small fast model costs a fraction of a cent per minute. A frontier model costs around $0.16 per minute, which is more than every other component combined.
Add the knowledge base, denoising and PII removal options and you are adding another two cents. Model choice is the entire cost story and nobody tells you that until you read the breakdown. Work out your per-minute figure with the model you actually intend to ship, multiply by your real call volume, and decide before you build rather than after. On learning resources, set expectations correctly.
There is no Retell academy, no course, no certification. There are reference docs, which are decent, a blog, a Discord and a community forum. If you arrive knowing how voice agents work and what you want yours to do, this is enough. If you arrive wanting to be taught the discipline, you will struggle, because the hard part of voice agents is not the API.
It is prompt design under interruption, handling the caller who answers a question you did not ask, deciding when to escalate, and building the analytics to know whether any of it worked. Retell gives you the machinery and assumes the judgement. My honest advice is to learn the concepts somewhere else and then come here to build. Three point nine.
Strong engineering, honest documentation, misleading headline pricing, and no teaching whatsoever.