Model tiers, monthly allowances and Auto

The platform's models now come in three tiers: Economy, Standard and Premium. Each has its own models, and each plan includes a monthly token allowance per tier, shared across the owner's workspaces.

When you start a session on the platform's models, the launch dialog lets you pick a tier. Each option shows what is left this month, or says "not in your plan" or "used up this month". A follow-up run keeps its parent's tier.

Auto picks a tier for each step. Planning, architecture, review and security steps use the best tier you have allowance for. The step that writes the code tries Standard first, and drafting steps try Economy first. The aim is results close to Premium while using far fewer Premium tokens.

The Usage page now opens with "AI allowance this month": a bar per tier, amber from 80% and red when used up. When a tier runs out, a session moves to your workspace's credits if you have some, or stops and tells you what ran out. It never switches tier by itself. (Update, 24 October 2026: running out of allowance still never switches tier. Separately, the platform administrator can now give a tier a fallback for when its provider doesn't answer at all; a call that was answered is never re-sent to another tier.) Sessions on your own provider key don't count against the allowance. More in models and cost controls.