Usage tier
The cap on how much you can use an AI subscription in a given period — usually measured in tokens processed, sometimes in messages or minutes.
A usage tier is the ceiling on your consumption within a billing cycle on an AI subscription. Most consumer AI products have one, though many hide it: you'll see terms like "rate limit" or "fair-use policy" instead of a specific number.
Why tiers exist: model inference costs the provider real money per request. A very heavy user can consume more than their subscription covers, so tiers align pricing to actual cost. Without tiers, subscriptions would need to be more expensive on average to cover heavy users.
Tiers can be measured in different units: - **Token-based** (LADLE, Anthropic API): counts input and output tokens over a rolling window. - **Message-based** (some competitors): fixed number of messages per day, regardless of length. - **Time-based** (some voice products): minutes of voice conversation per period.
Each has trade-offs. Token-based is the most honest about actual cost, but hardest for users to intuit. Message-based is easy to understand but doesn't reflect that a 40-message chat about a small thing is very different from 1 message with a 200-page PDF attached.
When you hit a usage tier, providers respond differently: some throttle (slower responses), some cut off, some silently downgrade to a cheaper model. The honest ones tell you which is happening.