Retell
A platform focused on voice AI agent orchestration, providing capabilities including conversational flow management, low-latency response, interruption detection, and telephony system integration. It is positioned as an integrated solution for upper-layer agent orchestration.
Core Facts
Timeline (last 90 days)
The Retell platform carries 100 million calls per month.
The test set and scoring criteria of Retell's voice agent benchmark test are not disclosed in the source material, making external reproduction difficult.
Deep reliance on a single model vendor reduces integration complexity, but it also means a high degree of dependence on a single vendor.
Retell is deeply tied to OpenAI's single-model system rather than widely diversified across multiple vendors.
The introduction of Codex helped Retell achieve an annualized revenue run rate of $1.6 million per employee.
GPT-6 Astra ranks first in Retell's voice agent benchmark test.
The operator of a small residential roofing company routed missed calls to an AI receptionist, using Retell configured with the GPT 5.6 Terra model, ElevenLabs' Jeremy voice, accurate transcription mode, dynamic response waiting, interrupt sensitivity set to 0.6, and real call latency of about 1.0 to 1.8 seconds.
The Retell platform offers three ways to structure a voice agent: single prompt, multi-prompt (state machine), and conversation flow (visual graph).
If the core requirement is conversation orchestration and call stability, a mature agent platform (such as Retell) may still hold engineering advantages; if voice naturalness is the biggest selling point and the team can handle conversation orchestration independently, the updated ElevenLabs may be worth considering.
Platforms like Retell add a conversation orchestration layer on top of the ASR, LLM, and TTS modules, responsible for managing state machines, invoking external tools, handling interruption logic, and managing integration with PSTN or VoIP.
3 more timeline events
All Facts (14)
传统语音AI普遍采用级联式架构，先用ASR将语音转文字，再交给语言模型处理，最后用TTS合成输出
75%UnverifiedThe test set and scoring criteria of Retell's voice agent benchmark test are not disclosed in the source material, making external reproduction difficult.
50%UnverifiedThe Retell platform carries 100 million calls per month.
50%UnverifiedGPT-6 Astra ranks first in Retell's voice agent benchmark test.
50%UnverifiedThe introduction of Codex helped Retell achieve an annualized revenue run rate of $1.6 million per employee.
50%UnverifiedRetell is deeply tied to OpenAI's single-model system rather than widely diversified across multiple vendors.
50%UnverifiedDeep reliance on a single model vendor reduces integration complexity, but it also means a high degree of dependence on a single vendor.
50%UnverifiedThe operator of a small residential roofing company routed missed calls to an AI receptionist, using Retell configured with the GPT 5.6 Terra model, ElevenLabs' Jeremy voice, accurate transcription mode, dynamic response waiting, interrupt sensitivity set to 0.6, and real call latency of about 1.0 to 1.8 seconds.
50%UnverifiedThe Retell platform offers three ways to structure a voice agent: single prompt, multi-prompt (state machine), and conversation flow (visual graph).
50%UnverifiedRetell is a platform focused on voice agent orchestration, with strengths in conversation flow management, low-latency response, and engineering optimization for call stability.
50%UnverifiedPlatforms like Retell add a conversation orchestration layer on top of the ASR, LLM, and TTS modules, responsible for managing state machines, invoking external tools, handling interruption logic, and managing integration with PSTN or VoIP.
50%UnverifiedIf the core requirement is conversation orchestration and call stability, a mature agent platform (such as Retell) may still hold engineering advantages; if voice naturalness is the biggest selling point and the team can handle conversation orchestration independently, the updated ElevenLabs may be worth considering.
50%UnverifiedFor the prototype validation stage, it is recommended to prioritize the free credits of all-in-one platforms like Vapi and Retell; for small-scale launch, switch to Telnyx or Plivo; and for large-scale production, choose Twilio or a self-built solution.
50%UnverifiedThe backend of GPT-6 Astra's voice mode calls Codex as the execution engine.
50%