Retell AI
An AI platform providing voice agent services, with built-in STT, TTS, telephony and LLM orchestration capabilities, and support for connecting custom models through a Custom LLM interface.
Timeline (last 90 days)
When using the Custom LLM approach, you must host the WebSocket server yourself on the call's critical path; the server should be as close to Retell's region as possible to control additional latency, and you should test it in practice before deployment.
Retell AI's voice agents bill large model calls by the minute by default, with frontier model prices between $0.045 and $0.16 per minute, stacked on top of the voice engine fee.
Retell AI provides a Custom LLM interface that allows opening a WebSocket to the user's server during a call, pushing real-time transcription text, and speaking back content streamed from the user.
Connecting an open-source model to the Retell Custom LLM slot requires only a bridge service of about 90 lines.
Retell's Custom LLM event flow follows: the server first sends a config event and the first sentence; after the caller finishes speaking, Retell sends a response_required with a response_id; the bridge invokes streaming completion and returns response events token by token; finally content_complete closes that turn.
Retell's response_id increases monotonically; when the caller interrupts, Retell sends a new response_required event carrying a larger response_id, and the bridge needs to stop pushing stale tokens from the old reply.
After creating an Agent in the Retell dashboard, select Custom LLM via Other options and set the Custom LLM URL to the wss:// bridge address; Retell will append the call id and establish the connection at the start of the call.
Retell's built-in LLM prices (marked date 2026-07-23): GPT 5.5 is $0.16/minute, GPT 5.4 or Claude 4.6 Sonnet is $0.08/minute, and GPT 4.1 is $0.045/minute.
Among Retell's other billing items, the voice engine is $0.055/minute, and with TTS and telephony fees unchanged, the savings from using Custom LLM occur only on the LLM line.
The Custom LLM slot only covers the LLM part; Retell's STT and TTS still go through its built-in providers, and third-party open-source TTS cannot be connected here; the original article mentions that Vapi provides an open TTS slot.
1 more timeline events
All Facts (12)
在全栈语音对话层，Vapi、Bland.ai、Retell AI等专注于语音代理的垂直平台正快速崛起
80%UnverifiedRetell AI's voice agents bill large model calls by the minute by default, with frontier model prices between $0.045 and $0.16 per minute, stacked on top of the voice engine fee.
50%UnverifiedRetell AI provides a Custom LLM interface that allows opening a WebSocket to the user's server during a call, pushing real-time transcription text, and speaking back content streamed from the user.
50%UnverifiedConnecting an open-source model to the Retell Custom LLM slot requires only a bridge service of about 90 lines.
50%UnverifiedRetell's Custom LLM event flow follows: the server first sends a config event and the first sentence; after the caller finishes speaking, Retell sends a response_required with a response_id; the bridge invokes streaming completion and returns response events token by token; finally content_complete closes that turn.
50%UnverifiedRetell's response_id increases monotonically; when the caller interrupts, Retell sends a new response_required event carrying a larger response_id, and the bridge needs to stop pushing stale tokens from the old reply.
50%UnverifiedAfter creating an Agent in the Retell dashboard, select Custom LLM via Other options and set the Custom LLM URL to the wss:// bridge address; Retell will append the call id and establish the connection at the start of the call.
50%UnverifiedRetell's built-in LLM prices (marked date 2026-07-23): GPT 5.5 is $0.16/minute, GPT 5.4 or Claude 4.6 Sonnet is $0.08/minute, and GPT 4.1 is $0.045/minute.
50%UnverifiedAmong Retell's other billing items, the voice engine is $0.055/minute, and with TTS and telephony fees unchanged, the savings from using Custom LLM occur only on the LLM line.
50%UnverifiedThe Custom LLM slot only covers the LLM part; Retell's STT and TTS still go through its built-in providers, and third-party open-source TTS cannot be connected here; the original article mentions that Vapi provides an open TTS slot.
50%UnverifiedWhen using the Custom LLM approach, you must host the WebSocket server yourself on the call's critical path; the server should be as close to Retell's region as possible to control additional latency, and you should test it in practice before deployment.
50%UnverifiedAll-in-one AI voice agent platforms such as Vapi, Retell AI, and Bland AI bundle telecom, STT, LLM, and TTS together, and typically offer free trial minutes.
50%