A committed spend agreement is the single biggest lever on your Anthropic bill, and the one buyers understand least. This guide explains how the commit bands work, where the overage and unused commitment traps sit, and how to size a commitment that serves you rather than the seller.
Anthropic, like most infrastructure sellers, rewards predictable revenue. When you commit to spend a set amount over a term, you earn a discount against the published rate. The larger and longer the commitment, the deeper the discount the seller can justify. That is the trade at the center of every enterprise Claude deal: you give up some flexibility, and in return you pay less per token than a buyer paying as they go.
The mechanics matter because each one is negotiable, and each one can quietly work against you if you do not set it deliberately. A commitment has a size, a term, a discount tied to a band, an overage rate for usage beyond the commit, and a treatment for commitment you do not draw. Get the size and the overage rate right and the deal protects you. Get them wrong and you have locked in a number you cannot hit or a penalty you did not see coming.
The rest of this guide walks each of those mechanics in turn, then points you to the deeper pieces on each one.
Discounts on Claude API spend tend to step up in bands rather than as a smooth curve. Crossing from one band into the next unlocks a materially better rate, which means the number just below a threshold is the most expensive place to sit. The illustrative bands below are directional, not a published price list, because Anthropic enterprise pricing is sales assisted and negotiated case by case. The pattern, though, is consistent.
| Annual commit band | Typical posture | What it tends to unlock |
|---|---|---|
| 250K to 1M | First serious commit | A meaningful discount off list and a named account team |
| 1M to 5M | Scaled production | Deeper discounting, price protection, and room on overage |
| 5M plus | Strategic account | Best rates, custom terms, and capacity guarantees worth negotiating |
The lesson for buyers is to know which side of a band you sit on before you negotiate. A company spending just under a threshold often does better by committing slightly more to cross it, provided the forecast supports the draw. A company that has been pushed above its real need is paying for a band it does not occupy. We benchmark where you actually sit and structure the commit to land in the right place.
Usage beyond the commit billed at the published rate rather than your negotiated rate. The most expensive default in the contract.
Commitment you do not draw often expires at the period boundary. No carryover, no credit, simply gone.
A commit set to the seller's ambition rather than your forecast. You pay for tokens you never consume.
A commit set too low to cross the band you belong in, leaving a better rate on the table.
A renewal that rebases your commit on grown usage, erasing the discount you won the first time.
A commit large enough that switching becomes impractical, weakening your leverage at the next negotiation.
The fastest way to improve a commitment is to need less of it. Before you size the number, reduce the demand underneath it. Routing work across Opus, Sonnet, and Haiku rather than running everything on Opus typically cuts aggregate spend 40 to 70 percent. Prompt caching returns up to 90 percent on the stable parts of your context. Batch processing runs the asynchronous jobs at 50 percent. We do that optimization first, then size the commit against the lower, real number. The savings are yours whether or not you renew, and the smaller commit is easier to hit and cheaper to carry.
"We thought our commit was as low as it could go. They optimized the spend underneath it, then resized the commitment and protected the overage rate. The bill stopped climbing."
Download the playbook, or have us run your commitment from forecast to executed order form.
Download playbookWeekly intelligence on Anthropic pricing moves and the buyer side counters that work.