Cheap Ticket Summaries: A $1.10 Cost Receipt

01 / Start with Gemini 2.5 Flash, then check the receipt

For routine support-ticket summaries, I would start by testing Gemini 2.5 Flash. That is a budget-first recommendation, not a claim that it wins every quality comparison. The job is narrow: preserve the customer's problem, troubleshooting already attempted, current status and next action. Pick the model that does that reliably at the lowest total cost.

Google lists standard paid text input at $0.30 per million tokens and output at $2.50 per million tokens, including thinking tokens. Those are Google's published rates, not a verified CheapRelay quote. Use them as a transparent reference calculation, then substitute the relay's displayed rates before buying. ([ai.google.dev](https://ai.google.dev/gemini-api/docs/pricing))

02 / Your receipt for 1,000 ticket summaries

Assume each request contains 2,000 ticket tokens plus 250 instruction tokens. Budget 170 output tokens for a compact handoff. These are workload assumptions, not measured averages. The calculation assumes no additional thinking tokens, cache usage, tools or retries.

• ■ Ticket text: 1,000 × 2,000 = 2,000,000 input tokens; reference charge: $0.60.
• ■ Instructions: 1,000 × 250 = 250,000 input tokens; reference charge: $0.075.
• ■ Summaries: 1,000 × 170 = 170,000 output tokens; reference charge: $0.425.
• ■ Total token charge: $1.10 for 1,000 tickets.
• ■ Unit cost: $0.00110 per ticket, or 0.11 cents.

The arithmetic uses Google's published standard rates. At the same assumed lengths, 10,000 summaries cost $11 and 100,000 cost $110. These are estimates for token processing, not complete operating bills or promised relay prices. ([ai.google.dev](https://ai.google.dev/gemini-api/docs/pricing))

03 / A longer summary adds $0.825 to the receipt

Keep the same input but let each summary grow from 170 to 500 output tokens. Across 1,000 tickets, output becomes 500,000 tokens and costs $1.25. Add $0.675 for input: the new total is $1.925, approximately $1.93. That is 75% above the original $1.10 estimate.

At these reference rates, an output token costs about 8.3 times an input token. That makes a compact handoff worth testing before spending time shaving a few words from your instructions. The savings calculation follows the published input and output prices. ([ai.google.dev](https://ai.google.dev/gemini-api/docs/pricing))

My proposed format is four fields: issue, actions tried, unresolved point and next owner. Ask for plain facts, no greeting and no repeated transcript. Treat 170 tokens as a planning target, not a guarantee; reject incomplete summaries rather than blindly truncating them.

04 / Thinking, caching and retries need separate lines

Visible summary length is not necessarily the whole output bill: Google's output price includes thinking tokens. An assumed extra 300 thinking tokens per ticket would add 300,000 billed output tokens across this workload, or $0.75 at the reference rate. Inspect usage records instead of estimating charges from the displayed answer alone. ([ai.google.dev](https://ai.google.dev/gemini-api/docs/pricing))

Do not pre-credit caching savings. Google's pricing lists cached input and cache storage separately; eligibility and actual usage need checking. CheapRelay bills real tokens, including cache reads and writes, so reconcile those categories against its own rates rather than importing Google's cache economics. ([ai.google.dev](https://ai.google.dev/gemini-api/docs/pricing))

Budget retries explicitly. If 50 of the original 1,000 requests require one identical rerun, that hypothetical 5% retry rate adds $0.055, bringing token spend to $1.155. This is a planning scenario, not an observed failure rate.

05 / Choose on cost per accepted handoff

Before rollout, test 100 representative tickets: short questions, long threads, reopened cases and ambiguous ownership. Have a reviewer check whether each summary preserves the problem, completed actions, unresolved status and next step. Mark invented resolutions as failures.

Track total token spend divided by accepted summaries. Also record correction time separately. For perspective, ten minutes of review at an assumed $30 hourly labour cost is $5—more than four times this example's initial API bill. Cheap generation is useful only when the handoff remains usable.

CheapRelay offers pay-as-you-go access with no monthly fee or subscription, which suits a small pilot before a larger commitment. Sign up for CheapRelay, check the current Gemini 2.5 Flash rates, and run that 100-ticket test. Keep the model if its accepted-summary cost and factual accuracy meet your requirements; reconsider it if corrections erase the token savings.