What AI models does Simple Chat use?

Simple Chat uses Anthropic's Claude Haiku 5.5 to write its replies and run its AI tasks, and Claude Opus 5.5 for the topics that your AI resolves least. The models are the same on both plans. You don't need your own API keys.

Customer replies

• Standard and Premium both use Claude Haiku 5.5.
• Easier and harder issue types and greetings use Claude Haiku 5.5, except in a low-resolution topic. Back-in-stock replies always use Claude Haiku 5.5.
• A low-resolution topic is a topic where your AI resolved 80% or less of your store's closed conversations in the last 7 days, as your own Analytics show. Replies in that topic use Claude Opus 5.5. When the rate goes above 80%, the topic goes back to Claude Haiku 5.5.
• Customer chat runs at high reasoning effort on both plans.

No reply falls back to another model. Simple Chat retries a failed request automatically. If it still fails, the conversation goes to your staff.

Staff and background tasks

The Premium staff-admin assistant uses Claude Haiku 5.5 with high reasoning effort. When it reviews an escalated conversation in a low-resolution topic, it uses Claude Opus 5.5 with high reasoning effort. The @SC staff assistant uses Claude Haiku 5.5 with extra-high reasoning effort, or Claude Opus 5.5 with high reasoning effort in a conversation of a low-resolution topic. Other tasks use their own reasoning effort, which doesn't depend on your plan. TypeSafe Jev helps classify requests, select relevant store information, and check reply quality for stores on both plans. Spam detection uses Jev too, with the OpenAI Decisions API only as a fallback for the spam decision.

You can change plans in Billing. Shopify asks you to approve the new per-resolution rate before the change takes effect.