Most products quote token limits as if tokens were a synonym for words. Inside Quantm, tokens are the budget the engine spends on its internal problem graph: how many data nodes can be allocated, how many inference steps can be recorded, how many independent verification paths can be pursued. Output length is one consumer of that budget, but it is rarely the dominant one. On a hard problem, the bulk of the spend goes into the derivation the user never sees.
This is why tier upgrades are not 'longer answers' — they are deeper answers. The free tier holds a small graph and produces a single derivation. Elite expands the graph and runs five independent derivations. Unstoppable expands further and runs twenty-five. The output-length difference between tiers is often modest. The verification-depth difference is dramatic. Users who measure the upgrade by output length miss the point of the upgrade.
The structural implication is that token spend on Quantm is a measurable commitment to accuracy, not to verbosity. A single-path derivation that is wrong costs the same as one that is right; a five-path derivation that catches the wrong-result candidate before output is worth dramatically more than five times the single-path price, because it removes the downstream cost of acting on a silent error. That is the actual economics, and it is why the tiers are priced the way they are.