ElevenLabs Eleven Realtime v1 Max

ElevenLabs Eleven Realtime v1 Max is a native speech-to-speech model that handles transcription, reasoning, and synthesis within one continuous audio stream, delivering a time-to-first-audio of approximately 570 milliseconds. The Max variant prioritizes multi-step reasoning and accuracy over the Mini variant, achieving an 87.1 percent tool-use success rate on structured benchmarks and 94.1 percent on the Full-Duplex-Bench interruption evaluation. The model accesses the full ElevenLabs voice library of over 17,000 voices covering more than 90 languages, and supports professional voice cloning through the standard ElevenLabs cloning pipeline. Session audio is billed at $0.08 per minute for connected time plus LLM token usage, with burst pricing at $0.16 per minute above the plan concurrency limit. Standard paid plans support 10 concurrent sessions, scaling to 40 on the Business tier. Eleven Realtime v1 Max is best suited for Companion App Developers who require consistent voice identity across long conversations, natural barge-in handling, and reliable tool invocations. The extensive pre-made voice library reduces time spent on voice casting, while the OpenAI Realtime-compatible API surface allows teams migrating from gpt-realtime to switch with minimal code changes.