How to cut your Claude Code bill by 40% without changing your workflow

Where a 40% figure comes from

The 40% in the title is a modelled figure, not a customer average - we do not publish customer outcomes. It is a plausible outcome of combining three specific techniques on an Opus-heavy workload - the illustration at the end shows the arithmetic for the first one. None of them involves writing less code or switching away from Claude. Your own number depends on your mix of tasks.

1. Model-tier picking

Many developers default to Opus for everything. But a large share of everyday calls - completions, refactors, explanations - may not need the flagship tier. Claude Sonnet 4.6 lists at 40% less per token than Claude Opus 4.8, and Claude Haiku 4.5 at 80% less. Check the quality on your own tasks before you switch.

The fix: look at your usage breakdown by task type. Summarization, code review, and simple generation often do not need the flagship tier. Reserve Opus for complex multi-file reasoning and architecture decisions.

2. Prompt caching

Cache-read tokens cost a fraction of fresh input tokens - 0.10x the input rate on Claude Opus 4.8 and Claude Sonnet 4.6. If your system prompt or context window includes the same large block of text across calls, caching removes most of the cost of that repeated input.

• -Structure prompts so the static portion comes first
• -Keep the variable portion at the end
• -Reuse sessions where possible to maximize cache hits

3. Prompt-length discipline

Longer prompts cost more - not just in input tokens, but in the output they generate. A 2,000-token prompt that could be 400 tokens is wasting 80% of your input spend and often producing longer, less focused responses.

• -Cut preamble and restated context the model already has
• -Use references ("see the function above") instead of pasting code twice
• -Set explicit max_tokens to cap runaway output

Measuring the impact

The savings compound. As an illustration: a team doing 10,000 calls a month at an average of $0.08 a call spends $800. Moving 60% of those calls to Claude Sonnet 4.6 cuts their token cost by 40%, which on its own takes the bill to about $610; caching the repeated context and tightening prompts takes it lower still. Measure before and after on your own traffic - that is the only number that counts.