Guides / gemini bill spike
Why a Gemini bill spikes when the prompt gets a little longer
A Gemini bill spike with no traffic change often means prompts crossed 200,000 tokens, where Google charges a long-context rate at roughly double. Crossing that line changes your rate as well as your volume, so a modest prompt increase lands as a large bill increase. Culpa, a local-first LLM cost, margin, and forecast ledger, prices every call from a dated price book and names the feature behind the change.
Why this happens
Most cost intuition assumes rates stay fixed while volume moves. Gemini breaks that at 200,000 input tokens, where a second pricing tier takes over at roughly twice the rate. A retrieval change that raises average prompt size by a third can therefore raise cost by more than a hundred and fifty percent, because part of the traffic changes tier. Nothing in a token-count dashboard shows the boundary, so the spike reads as unexplained until somebody plots the distribution against 200k rather than the average.
What this usually looks like
- Average prompt size grew modestly and Gemini spend grew by multiples of that.
- A retrieval or context change shipped shortly before the increase started.
- Some calls sit above 200,000 input tokens and nobody has counted what share.
- Your own cost model reconciles for short prompts and reads low for long ones.
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 |
|---|---|---|
| Modelling Gemini cost with a single input rate. | A single rate is right below 200k and roughly half the truth above it, so long calls read cheap. | Model two tiers and track what share of calls sits above the boundary. |
| Watching average prompt length rather than the distribution. | An average well under 200k can hide a tail sitting above it, and the tail carries the tier change. | Plot the histogram against the boundary. The share above the line is the number that matters. |
| Letting retrieval decide prompt size without a ceiling. | Retrieval expands to fill whatever budget exists, so the boundary gets crossed by accident. | Cap assembled context below the boundary deliberately, and measure the quality cost of that cap. |
Run this check tonight
- Count what share of your Gemini calls exceeded 200,000 input tokens last month.
- Check whether your internal cost model uses one input rate or two.
- Find the release that changed retrieval or context assembly, and line it up against the spend curve.
- Compute the cost of the same workload capped just under the boundary, and compare.
- Confirm your longest-prompt feature really is the one you assume, by measuring rather than by memory.
A 33% longer prompt that cost 157% more
Illustrative example
A document feature on Gemini 2.5 Pro. Below 200k the price book rate is $0.00125 per 1k input and $0.010 output. Above 200k, Google's pricing page lists roughly double, $2.50 and $15.00 per million. Volumes are modelled.
A third more context produced two and a half times the bill, because the extra tokens crossed a tier boundary that no token dashboard draws.
Every number, with its confidence and source
| Figure | What it means | Confidence | Source |
|---|---|---|---|
| $0.245 | modelled cost of one 180k-token call on Gemini 2.5 Pro below the long-context boundary | calculated | (180 x $0.00125) + (2 x $0.010) per 1k, using real Gemini 2.5 Pro rates from the price book, effective 2026-07-02. Token volumes are modelled. |
| $0.245 to $0.630 | modelled cost per call either side of the 200,000-token tier boundary | estimated | The upper endpoint uses the above-200k tier listed on the Gemini pricing page, fetched 2026-08-01, which the price book deliberately doesn't hold. A range because token volumes are modelled. |
What a generic answer can’t know
The provider applies the tier correctly and tells you the total. Which feature assembled the long prompts, what share of calls crossed the line, and which release moved them there are facts about your own retrieval code. Culpa records them on your infrastructure, keeps your prompts and responses there, and counts the calls to run your plan.
Questions founders ask next
Does Gemini charge more for long context?
Yes. Google's pricing page lists a second tier above 200,000 input tokens at roughly double the rate. Culpa's price book deliberately seeds the at-or-below-200k tier, which migration 0009 documents as a cost floor rather than a full model of the ladder.
How do I tell whether the boundary caused my spike?
Count the share of calls above 200,000 input tokens before and after the increase. If that share moved, the tier change explains part of the spike, and a token-count average will never show it on its own.
Is capping context below 200k worth it?
Often, because the marginal tokens cost double and the last few retrieved chunks usually contribute least. Measure the quality difference at a cap rather than assuming it, then decide with both numbers in front of you.
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: Gemini API pricing. Last reviewed 2026-08-01. Plain text version.