ElevenLabs has introduced its own hosted large language models (LLMs) within their agents platform to enhance the performance, efficiency, and cost-effectiveness of voice agents. By integrating open-source models directly into their infrastructure, ElevenLabs provides ultra-low latency and reduced reasoning costs, allowing for the deployment of voice agents without the need for external providers. The platform features GLM 4.5 Air for high-level reasoning and Qwen3-30b-a3b for quick, natural dialogues, with both models offering significant cost advantages over alternatives. The co-located architecture combines these LLMs with ElevenLabs' proprietary Speech to Text, Text to Speech, and turn-taking models, improving latency, reliability, and data security. The ElevenLabs platform also includes tools for designing conversation flows and testing agent reliability, integrating seamlessly into CI/CD workflows to ensure robust and compliant deployments.