Model Overview
RouterBus is an API relay: models from several upstream providers sit behind one address and one API key.
What is available
The models currently connected come from three product lines:
- Anthropic Claude — Sonnet and Opus, including thinking variants
- OpenAI GPT / Codex — GPT chat models, code review models, and GPT image models
- Google Gemini — Gemini chat models and image models
Upstream providers add, rename and retire models constantly, so this page deliberately does not list them.
Model Square is the source of truth
Model Square is the only accurate list: it reads the same configuration the billing engine does, and shows which models are actually available right now, their live prices, and which endpoints each one accepts.
Once you are signed in, the API Keys page also shows the models a given key can reach.
Model IDs
The model ID you pass to the API is exactly the string shown in Model Square — for example claude-sonnet-4-6.
Do not type model IDs from memory, and do not reuse naming from other platforms (such as older date-suffixed versions). Copying from Model Square is the only way to get it right.
Endpoints
A model may be open on more than one endpoint; the model's detail page in Model Square says which:
| Endpoint | Address | Clients |
|---|---|---|
| OpenAI-compatible | https://www.routerbus.com/v1 | openai-python, openai-node, Codex CLI, and most third-party clients |
| Claude native | https://www.routerbus.com | Claude Code, Anthropic SDKs |
| Gemini native | https://www.routerbus.com/v1beta | Gemini CLI, Google GenAI SDKs |
The OpenAI-compatible endpoint has the widest coverage: nearly every model is open on it, and it is the easiest way to switch models without touching your code.
Choosing a model
- Everyday coding and chat — the Claude Sonnet line, the most balanced option
- Hard reasoning, architecture, long-running tasks — Opus and flagship GPT models, or a thinking variant
- Quick questions and batch work — Gemini flash and other lightweight models, lower latency and lower cost
- Image generation — the GPT and Gemini image models
If you are unsure, get the pipeline working with a Sonnet or flash model first, then move up if the output is not good enough.
Groups
Models are published in groups, and you can pin an API key to a group when you create it. The default is automatic group selection, which picks a group by availability. Model Square shows which models each group serves.