Baseten launches Voice Agent Stack and Realtime Voice in private preview
Start here
Baseten launched the Voice Agent Stack, infrastructure for running production voice agents on its Dedicated Inference, and Realtime Voice, a reference voice agent API built on it. Both are in private preview and require joining a waitlist. Realtime Voice chains Baseten-optimized open-weight speech models with a Pipecat-based orchestrator. Baseten says it scores 93% on Full-Duplex-Bench and 73.5% on τ²-Bench Voice, and costs about $0.012 per audio minute at 32 concurrent streams, or an estimated $0.008 per call-minute at 56 concurrent sessions including model API delegation. Calls delegated to larger reasoning models are billed per token like standard LLM endpoints. The stack keeps the orchestrator and models in one cluster, scales each component independently, supports post-training at each stage, and uses the OpenAI Realtime protocol to ease migration.
Generated from reports · updated 1 h ago
Report timeline
Follow the reports to see the event from different sides.
- Baseten BlogBaseten launches Voice Agent Stack and Realtime Voice voice agent API in private preview
Baseten launched the Baseten Voice Agent Stack, infrastructure for running production voice agents on Baseten Dedicated Inference, together with Baseten Realtime Voice, a reference voice agent API built on it; both are in private preview. Realtime Voice runs a cascade of Baseten-optimized open-weight speech models with a Pipecat-based orchestrator, costs about $0.012 per audio minute at 32 concurrent streams, and reports 93% on Full-Duplex-Bench and 73.5% on τ²-Bench Voice. The stack keeps the orchestrator and models in one cluster, scales each component independently, supports post-training at each stage, and speaks the OpenAI Realtime protocol for easier migration.
Heat for this event
Not enough continuous observations yet to draw a trend.