You can now use the complete GPT-5 family (GPT-5, GPT-5-mini, GPT-5-nano), as well as the gpt-oss model, which runs on Groq, as the LLM for any agent on Vapi
Why It Matters for Voice Agents
We’re making 3 variants available:
GPT-5: The new default for balanced performance.
GPT-5-Mini: A faster variant for high-volume tasks where cost and speed are more critical than complex reasoning.
GPT-5 Nano: For workflows where speed is the absolute priority.
GPT-5 uses a router to dynamically switch between a fast model for simple turns and a "thinking" model for complex reasoning. For voice, this means:
Dynamic Reasoning: Low latency for simple turns, deep thinking for complex ones. All in the same call.
Fewer Hallucinations: More trustworthy agents for regulated domains like healthcare and finance.
Better Instruction Following: Cleaner, more reliable tool calls with less prompt engineering.
Less Sycophancy: More direct, natural-sounding conversations that don't feel like talking to a generic bot.
Where It’s Useful
We’re already seeing developers use it for:
Customer support agents that handle both quick chats and complex, multi-step troubleshooting.
Outbound sales agents that need to reason through complex product details on the fly.
Any workflow where an agent must reliably execute a series of tool calls without error.
Real Takeaway
Vapi’s job is to give you the best components. GPT-5’s new architecture is a major step forward for building production-grade voice agents.
But our platform remains open. Test it against other models.
The best model is the one that works well with your use case.
