What it is
OpenAI sells API access to its GPT models. Most agents here call it natively, and its chat-completions format is the common ground of local servers and gateways.
Strengths
- Supported natively by OpenClaw, Hermes Agent, Nanobot and ZeroClaw
- Ollama, vLLM and OpenRouter expose the same API shape, so switching between them is a base URL
Limits
- Models are retired on a schedule: pin the model you use and watch the deprecations page
- Prompts leave your hardware, as with any hosted model
Best for
- Teams already on OpenAI for other work
- A second provider to fall back on
Not for
- Data that must stay on your hardware (use a local model)
Our take
Our opinion: worth configuring even if it is not your main provider, because an agent that can fall back to a second provider keeps working through an outage or a rate limit. Pin the exact model name in the agent's config: a model alias that moves to a new release changes the agent's behaviour without warning.
Sources
Read on 2026-10-27.
Alternatives
- Anthropic (Claude API): Anthropic's API for the Claude models, which several of the agents we follow were built around.
- OpenRouter: One OpenAI-compatible API in front of many model providers, with fallback between them.