LLM models for voice AI
The reasoning brain — the model that understands the caller and decides what to say.
Supported providers
OpenAI
Industry-leading reasoning, broad tool-call support. Only low-latency chat models are listed — slow reasoning models (o1 / o3) are deliberately excluded, they stall the first reply on a live call.Anthropic
Strong reasoning and tool use; great for nuanced personas. For live calls prefer Haiku (fastest/cheapest) or Sonnet (balanced) — Opus is higher quality but pricier and slower to first token.Groq
Fastest tokens-per-second on hosted open models.Google Gemini
Gemini 2.5 family via the Gemini API (OpenAI-compatible). Flash and Flash-Lite are fast + cheap with huge context — strong defaults for live calls; Pro for the hardest reasoning.xAI Grok
Grok via the xAI API (OpenAI-compatible). Grok 4 Fast (non-reasoning) is built for low-latency chat with a 2M context; Grok 3 mini is the budget pick. Always-on reasoning Grok 4 is excluded — too slow to first token for live calls.OpenRouter
One key, hundreds of models through an OpenAI-compatible gateway. Type or paste ANY model ID from openrouter.ai/models (e.g. anthropic/claude-sonnet-4.5) — the suggestions are just popular call-friendly picks; pricing roughly tracks the underlying provider.Azure OpenAI
OpenAI models hosted in your Azure tenancy. The deployment you configure below selects the model; the chosen model here is used for the cost/latency estimate. Requires endpoint + deployment + API version.AWS Bedrock
Claude, Llama and Amazon Nova through AWS Bedrock (unified ConverseStream API). Bring your own AWS credentials, or leave them blank to use the platform AWS account.Custom API (full agentic workflow)
You already built an agentic workflow on your server — tools, RAG, memory, system prompt all live with you. We just attach voice (STT + TTS) and pipe each user turn through your SSE endpoint.Custom LLM (open-source / self-hosted)
Point at any OpenAI-compatible endpoint — Ollama, vLLM, llama.cpp, LM Studio, or a self-deployed model. We treat it exactly like OpenAI: same chat-completions wire format, your model + your URL. Tools work too. ₹300.00 free credit on signup
Build a voice agent with llm models
Sign up free and get ₹300.00 in credit — no card required. Connect your number, pick a template, and go live in minutes.