Coding Assistant Habits That Double Your Bill

01 / Your $0.60 session has a $1.20 version

A coding assistant bill rarely arrives labeled “unnecessary repository dump” or “explanation nobody read.” It arrives as tokens. That makes expensive habits easy to miss: each request looks small, while repeated context and oversized answers quietly increase the total.

Here is a receipt you can audit. Every figure below uses illustrative rates of $2 per million input tokens and $10 per million output tokens. These are arithmetic assumptions, not CheapRelay prices, a model quote, or a benchmark. The example assumes no cache reads or writes; those need separate accounting when present.

02 / Receipt A: twenty focused requests cost $0.60

Imagine a coding session with 20 requests. Each sends 10,000 input tokens and receives 1,000 output tokens. Those are assumed lengths, not measured averages. The input includes the instructions and relevant code; the output contains a bounded patch and a short explanation.

• ■ Input: 20 × 10,000 = 200,000 tokens. At $2 per million: $0.40.
• ■ Output: 20 × 1,000 = 20,000 tokens. At $10 per million: $0.20.
• ■ Session total: $0.60.

This is our comparison receipt, not a promise that every task fits those limits. A migration or difficult debugging session may genuinely need more context and longer answers.

03 / Habit one: doubling context adds $0.40

Now assume every request sends 20,000 input tokens instead of 10,000. Perhaps you paste an entire log rather than the failing section, include unrelated files, or keep carrying an obsolete implementation through subsequent requests.

• ■ New input: 20 × 20,000 = 400,000 tokens.
• ■ New input charge: $0.80.
• ■ Extra charge versus Receipt A: $0.40.
• ■ Running total, with output unchanged: $1.00.

The expensive habit is not providing context. It is sending context without checking whether it changes the answer. Before another request, remove resolved errors, duplicated snippets, and files unrelated to the current dependency path. Keep the failing test, relevant interfaces, and constraints.

04 / Habit two: doubling answers adds another $0.20

Suppose each response also grows from 1,000 to 2,000 output tokens. Asking for a full-file rewrite, a detailed walkthrough, and several alternatives can produce more material than you need to apply one change.

• ■ New output: 20 × 2,000 = 40,000 tokens.
• ■ New output charge: $0.40.
• ■ Extra output charge versus Receipt A: $0.20.
• ■ Receipt B total: $0.80 input + $0.40 output = $1.20.

That is exactly twice the original $0.60, with the request count still at 20. This example demonstrates billing arithmetic, not equal code quality or equal task completion. A shorter answer is only a saving if it remains sufficient.

Try a specific output contract: return the patch, identify the affected tests, and explain only non-obvious decisions. Ask for full files when you actually need full files.

05 / Five unnecessary reruns push the receipt to $1.50

At Receipt B’s assumed lengths, one additional request costs $0.06: $0.04 for input and $0.02 for output. Five unnecessary reruns add $0.30, bringing the session to $1.50, or 2.5 times the starting bill.

Before retrying, identify what changed. Supply the actual error, clarify an acceptance criterion, or narrow the requested edit. A retry that adds no new evidence deserves scrutiny, especially when it resends the same large context.

06 / Keep separate lines for input, output, and cache

CheapRelay offers pay-as-you-go access through one key, with no monthly fee or subscription. Billing follows real token usage, including cache reads and writes. That makes a receipt mindset useful: compare each billing category rather than treating every token as the same expense.

Sign up for CheapRelay and audit one real coding session: record its billed token categories, remove unnecessary context, and tighten the requested output. Compare costs alongside completed work. The goal is not the shortest prompt; it is paying for information and answers you actually use.