Skip to main content
ShingleAI agents run through OpenRouter, which exposes a unified API over Anthropic, OpenAI, and Google models.

Available models

ModelTypical use
Claude Sonnet 4.6 (default)Balanced reasoning + cost for most agent tasks
Claude Haiku 4.5Fast, cheap — great for simple classification or short replies
Claude Opus 4Hardest reasoning, highest cost
GPT-4oStrong general model with broad tool use
GPT-4o MiniSmaller GPT-4o for cost-sensitive work
Gemini FlashLow-latency Google model
Gemini ProHigher-capability Google model
No tier gating today — every organization can select any of these models. Usage is billed by the token, converted into ShingleAI credits. See Admin Guide → Billing for how credits work.

Picking a model

Rough guidance:
  • Default to Claude Sonnet 4.6. It hits the best balance of quality and cost for most agent workloads.
  • Move to Haiku or Flash if latency and per-call cost matter more than reasoning depth (for example, a classifier agent that only picks a label).
  • Move to Opus only when a task consistently fails at Sonnet — the cost delta is substantial.

Changing the model

Open the agent’s detail page and edit the Model field. The change applies to all new conversations; conversations already in progress continue on whichever model they started.

Next steps

Tool permissions

Control what the agent can do

Chat with an agent

The chat UI