Guides / model upgrade cost increase
What a model upgrade does to an unchanged workload
A model upgrade cost increase is the same workload priced at a new rate, and it happens whether or not you changed anything. Claude Sonnet 5's introductory pricing ends on 2026-09-01 and rises 50%. Culpa, a local-first LLM cost, margin, and forecast ledger, prices every call from a dated price book, so a rate change lands as a forecast rather than as an invoice.
Why this happens
Two different events get called a model upgrade. One is a decision: you move traffic to a newer model and accept a different rate. The other is a schedule: the rate on the model you already run changes on a date the provider set. Both produce the same arithmetic and only the first appears in anybody's plan. Because rate changes are dated, they're the one class of cost increase that's entirely forecastable in advance, and they're still the one that surprises teams most.
What this usually looks like
- A dated rate change is approaching on a model you run and nothing in your forecast reflects it.
- A model migration was decided on quality alone, with cost checked afterwards.
- Your cost model holds one rate per model rather than a rate with an effective date.
- Nobody knows which models in production have a scheduled price change.
Free, no card, no account
Run the free Cost Leak Scan
It shows your most expensive conversation before you install anything.
Mistakes that cost the most
| Mistake | Why it hurts | Do instead |
|---|---|---|
| Holding one rate per model in your cost model. | A model can carry several rates over time, so a single rate silently mis-prices every period but one. | Store rates with effective dates and price each call at the rate in force when it was made. |
| Migrating models on quality alone. | A frontier model at five times the rate can cost more than the whole feature earns. | Price your actual monthly volume at both rates before the migration, then decide with both numbers. |
| Treating an introductory rate as the real one. | Introductory pricing has an end date, so a forecast built on it under-reads from that date onward. | Forecast at the post-introductory rate, and treat the discount as a temporary saving rather than a baseline. |
Run this check tonight
- List every model in production and check whether any carries a dated rate change.
- Reprice last month's volume at the new rate. That difference is already scheduled.
- Confirm your cost model prices calls by the rate in force on the call date rather than by today's rate.
- For any migration under discussion, compute the same workload on both models before the quality debate.
- Check what the same workload would cost on the tier below, so the upgrade has something to be compared against.
One workload, four rates
Illustrative example
A monthly workload of 900 million input tokens and 140 million output tokens. All four rates below are real price book rows, quoted per million. The volumes are modelled.
A 50% rise arrives on a known date with nobody deciding anything. The increase alone equals the full cost of running the same work on the tier below, which makes the September date a decision point rather than an event.
Every number, with its confidence and source
| Figure | What it means | Confidence | Source |
|---|---|---|---|
| $3,200 to $4,800 | modelled monthly cost of one unchanged workload on Claude Sonnet 5, before and after 2026-09-01 | estimated | Both endpoints use real Claude Sonnet 5 rows from the price book, $2.00 and $10.00 per million effective 2026-07-02 and $3.00 and $15.00 effective 2026-09-01. A range because the token volumes are modelled. |
| $1,600 | modelled monthly increase from the scheduled Sonnet 5 rate change, at unchanged volume | calculated | $4,800 minus $3,200, from the teardown arithmetic at real price book rates. |
What a generic answer can’t know
The rate change is public. What it costs you depends on your own token volumes per model, which no provider page can tell you and no single blended figure will approximate. Culpa prices your real volumes against a dated price book on your infrastructure, keeps your prompts and responses there, and counts the calls to run your plan.
Questions founders ask next
When does Claude Sonnet 5 pricing change?
On 2026-09-01. Anthropic's pricing page states that introductory pricing of $2 and $10 per million input and output runs through August 31, 2026, after which standard pricing of $3 and $15 per million takes effect, a 50% rise.
How do I forecast a dated rate change?
Take last month's token volumes for that model and multiply by both rates. The difference is your scheduled increase, and because it's dated rather than probabilistic it belongs in the budget rather than in a risk register.
Is upgrading to a newer model usually cheaper?
Not reliably. Newer models sometimes launch below the ones they replace and sometimes above, and introductory rates expire. The only dependable method is pricing your own volumes at both rates, because the direction changes model by model.
On your infrastructure
Culpa runs on your infrastructure. Your prompts and responses never leave it. Culpa counts calls to run your plan, and it fails open, so if it ever breaks your app keeps running.
How Culpa works
Find the culprit. Not just the total.
Your dashboard shows what you spent. It stops short of who spent it. Culpa shows the conversation, the user and the feature behind it.
Your prompts stay local.
Culpa runs on your own infrastructure. What you send to a model reaches us at no point.
Every dollar has a name.
Follow any charge to the conversation, the user, the feature and the customer behind it.
See the bill before it lands.
Cost your next feature before you ship it. You get the likely bill and the worst case, at best, median, p90 and p99.
Three steps to your first answer.
Change one base URL.
Or drop in the Python or TypeScript library.
Find your most expensive conversation.
In the first session, not the first week.
Cost your next feature before you ship it.
Why the bill went up
Example dashboardCalls traced
418,209
across 3 projects
Spend this week
$378.41
+ $182 vs last week
Failed calls
312
74% retried, and you paid for all of them
+ $182 this week traced to one culprit
Spend over 14 days
Most expensive users
Next week forecast
Graded against reality. Accuracy shown as results land.
Free, no card, no account
Run the free Cost Leak Scan
It shows your most expensive conversation before you install anything.
Keep reading
Sources: Anthropic pricing. Last reviewed 2026-08-01, rates effective 2026-07-02. Plain text version.