Configure model routing in LLM Proxy

Route model calls through one governed path

LLM Proxy gives model traffic a consistent route. That route can record provider, latency, token usage, cost, policy decisions, and refusal context before a response becomes part of an agent run.

Supported OpenAI models

OpenAI supports general reasoning, coding, realtime voice, transcription, and embeddings through the managed catalog.

• General and chat — GPT-5.2, GPT-5.1, GPT-5, GPT-5 mini, GPT-5 nano, GPT-5.2 Chat*, GPT-5.1 Chat*, and GPT-5 Chat*.
• Coding — GPT-5.2 Codex, GPT-5.1 Codex Max, GPT-5.1 Codex, and GPT-5 Codex.
• Higher-compute reasoning — GPT-5.2 pro and GPT-5 pro.
• Earlier general models — GPT-4.1, GPT-4.1 mini, GPT-4.1 nano, GPT-4o, and GPT-4o mini.
• Voice and transcription — GPT Realtime 2.1 and GPT Realtime Whisper.
• Embeddings — Text Embedding 3 Small.

* Chat-labeled models are supported, but are not recommended for long-running fully autonomous agents. Prefer a general or reasoning model for autonomous loops. The Chat label is not a requirement for Iron Gorilla chat: supported non-Chat models work with the chat feature too.

Supported Anthropic models

• Claude Opus 4.8
• Claude Sonnet 4.6
• Claude Haiku 4.5

Supported xAI models

Grok 4.5 adds xAI as another provider in the existing proxy. It supports chat and tool use with a large context window. It appears only when the xAI provider and model are enabled for your organization.

Choose preferred and fallback pairs

Review the enabled catalog, then choose preferred and fallback provider-model pairs for the default route or a narrower team, agent, or environment scope. Existing routes without a model choice continue to follow the request model.

Availability depends on provider enablement, model enablement, and organization policy; the catalog in LLM Proxy is the source of truth for your workspace. Apply policies before requests leave the boundary, and use an LLM trace to explain one call.