A Redditor's Sonnet 5 switch erased their Max rate limits

A Reddit user switched to Sonnet 5 and watched Max-subscription rate limits vanish, per a 123-upvote post in r/ClaudeAI. They hadn't touched the model in three months before making the switch.

On a fixed-price Max plan, the constraint is how fast you burn quota, not per-token cost. That's a different optimization problem than the one pricing coverage usually solves. It flips the usual advice to save the expensive model for hard problems.

Cheap tiers type, one tier reviews

A GitHub project called sol-advisor picked up 853 stars in two days doing the same math for OpenAI's GPT-5.6 lineup. Implementation work runs in the cheap Luna and Terra tiers. Then a Sol-tier review, started with no prior context, has to sign off before anything ships. The review is a mandatory gate, per the repo.

That structure doesn't need GPT-5.6 or Claude specifically. Any provider with tiered models and a subscription cap creates the same incentive: do the typing cheap, spend the expensive model's context on judgment.

Quota burn rate, not per-token price, is becoming the variable that decides which model builders default to. Watch whether Anthropic or OpenAI start routing requests across tiers automatically. Developers are already hand-rolling the split themselves, in a Reddit thread and a two-day-old GitHub repo. Patterns that spread this fast tend to get productized.