When Anthropic offers you a discount for committing to a level of API spend, the number sounds like a fixed schedule. It is not. The commit tiers in 2026 are bands of discretion, ranges within which the account team can move depending on how large your commitment is, how long you will sign for, and how the negotiation goes. If you treat the first percentage you are shown as the going rate, you are accepting an opening position. This is a buyer side guide to how those bands actually work and how to size a commitment you will use.
We negotiate with Anthropic and study nothing else, so we read these tiers the way the account team reads them. The goal here is to give a procurement leader and an engineering leader the same picture, so the two of you can decide together rather than guessing.
A committed spend agreement is a promise. You agree to spend at least a set amount on the Claude API over a term, usually a year, and in exchange Anthropic gives you a lower effective rate than pay as you go pricing. The larger the promise, the deeper the discount. That is the whole mechanism, and it is genuinely valuable when the commitment matches real usage. The danger is in the second half of the trade, which the headline discount tends to hide.
The discount applies to the rate. The commitment applies to the volume. A buyer who fixates on the rate and ignores the volume can sign a deal that looks like a saving and behaves like a penalty, because a discount on spend you never reach is not a discount at all.
The bands rise in steps as the committed amount grows. A modest commitment earns a modest discount, because it is not large enough to justify deep concessions and the representative has little room to move. As the commitment climbs into the higher tiers, the discount widens, because a larger guaranteed spend is worth more to Anthropic and the account team gains the authority to give more away to win it.
The practical lesson is that the per token rate you can reach depends on the size of the commitment you are willing to make. Comparing your offered discount to a peer at a different commitment level tells you nothing, because you are not buying the same thing. What matters is where your real annual usage sits, and which band that usage naturally lands in once you have optimized it.
The single most expensive mistake we see is committing to a tier based on current, unoptimized usage. Most enterprises arrive at the negotiation running a high share of traffic on the most expensive model, with no caching and no batching. Their consumption looks large, so they commit to a large tier to earn the headline discount. Then they optimize, usage drops, and they are locked into a commitment they cannot fill.
The order should be reversed. Model the spend you will have after routing across Opus, Sonnet, and Haiku, after prompt caching, and after batch where it fits. That optimized number, not today's inflated one, is the figure you should size the commitment against. For most buyers the optimized aggregate sits forty to seventy percent below uniform use of the top model, which can move you down a tier and still leave you comfortably inside it.
Four things move the discount you can reach, and a prepared buyer uses all of them deliberately.
Every commit agreement needs an answer to a simple question: what happens if you spend more than you committed. The good answer is that overage is billed at your committed, discounted rate, so growth stays cheap. The bad answer, and a common default, is that overage reverts to standard pay as you go pricing, which means the moment you exceed the commitment your marginal cost jumps back to list. A growing workload can quietly spend a large share of its tokens at the worse rate without anyone noticing until the invoice arrives.
Negotiate the overage rate as carefully as the headline discount. A commitment with cheap overage protects you on the upside. A commitment with expensive overage punishes the exact success the tool is supposed to deliver.
The mirror image of overage is shortfall. If you commit to a tier and do not reach it, what happens to the gap. In many agreements the unused commitment simply disappears at the end of the term, which means you paid for tokens you never used. Some buyers can negotiate a rollover, where unused commitment carries into the next period, or a true forward that converts the shortfall into future credit rather than lost spend. The default, again, favors the vendor, so silence on this point should be read as a cost.
Pull the threads together and the choice becomes a disciplined exercise rather than a guess. Start with your optimized annual forecast, the spend you will have after routing, caching, and batch. Choose the tier that your realistic usage will fill comfortably, not the one whose headline discount looks largest. Pair that tier with overage at the committed rate so growth stays protected, and with a treatment of unused commitment that does not simply burn the gap. Lock the rate across the term if you are signing for more than a year. Time the signature to a moment when Anthropic has a reason to move.
A commitment built that way earns a real discount on spend you will actually incur, protects you when usage rises, and does not penalize you when it does not. That is the difference between a tier that saves money and a tier that merely looks like it does.
Picture a buyer whose current Claude spend runs hot because almost everything is on the top model in real time. On paper they look like a candidate for one of the higher commit tiers, and the account team will happily size the deal there, because a larger commitment locks in more revenue. If that buyer signs at the tier their raw usage implies, then optimizes, they will spend the next year well below their commitment, paying for tokens that never get used.
Now run it the other way. The same buyer routes traffic across Opus, Sonnet, and Haiku, caches the repeated context that grounds their workloads, and moves their overnight evaluation runs into batch. Their aggregate spend falls sharply, often by half or more. The optimized figure lands them a tier lower, and at that lower tier the commitment is one they will fill comfortably with room to spare. They still earn a real discount, but on spend they will genuinely incur, and they carry none of the shortfall risk that the higher tier would have created. The lesson is not that lower tiers are better, it is that the right tier is the one your optimized usage fills, and you can only know that after you have done the optimization.
A commit tier lives or dies on the language around it, not on the percentage in the headline. Before you sign, make sure the agreement is explicit on a short list of points that vendors often leave silent, because silence in a contract defaults to the party that wrote it.
Each of these is ordinary to ask for and quietly expensive to omit. A buyer who reads the tier as a complete instrument, the rate plus all of its terms, signs something defensible. A buyer who reads only the discount signs something that looks the same on the page and behaves very differently on the invoice.
Get a quote for a bounded engagement. Fixed fee or gainshare, no risk to you.
Get a QuoteWeekly intelligence on Anthropic pricing moves and the buyer side counters that work.