How much does Claude Code cost per engineer per month?
Claude Code cost per engineer depends on the billing plan, model and token mix. The worked API-cost examples below come to about $5, $80 and $770 a month on Sonnet 5. They are illustrative scenarios, not a measured average, median or forecast for your team. Every token assumption is listed so you can substitute your own usage.
What are the token rates?
The table uses Anthropic's standard API rates for these specific models, checked on 5 September 2026, in US dollars per million tokens. Cache writes here use the five-minute tier. One-hour writes, batch pricing, fast mode and negotiated terms need their own rates.
| Model | Fresh input | Output | 5-minute cache write | Cache read |
|---|---|---|---|---|
| Haiku 4.5 | $1.00 | $5.00 | $1.25 | $0.10 |
| Sonnet 5 | $2.00 | $10.00 | $2.50 | $0.20 |
| Opus 5 | $5.00 | $25.00 | $6.25 | $0.50 |
| Fable 5 | $10.00 | $50.00 | $12.50 | $1.00 |
These rows deliberately name model versions. A newer model can have different cache pricing, so do not apply a multiplier to an unlisted version. The usage-log guide explains which counters feed the calculation.
What usage does each example assume?
The three scenarios below are made up. Light, moderate and heavy are labels for these inputs, not percentiles from a population of engineers. An agent-day means one day with the usage shown; inactive days contribute zero.
| Scenario | Fresh input | 5-minute cache write | Cache read | Output | Agent-days/month |
|---|---|---|---|---|---|
| Light | 2,500 | 50,000 | 900,000 | 25,000 | 8 |
| Moderate | 20,000 | 600,000 | 9,000,000 | 110,000 | 18 |
| Heavy | 320,000 | 4,800,000 | 80,000,000 | 800,000 | 21 |
For each category, divide tokens by one million and multiply by that category's rate. Sum the four amounts to get a daily estimate, then multiply by agent-days. The moderate Sonnet 5 example is $0.04 fresh input + $1.50 cache writes + $1.80 cache reads + $1.10 output = $4.44 a day, or $79.92 across 18 days.
The heavy example is $0.64 + $12.00 + $16.00 + $8.00 = $36.64 a day. Its cache reads are about 93% of tokens, but less than half the estimated cost. Counting all tokens at the fresh-input rate would miss that distinction.
What do the scenarios cost per month?
These are the exact totals for the invented inputs above, rounded to cents. The introductory $5, $80 and $770 figures are approximate labels for the Sonnet column.
| Scenario | Haiku 4.5 | Sonnet 5 | Opus 5 | Fable 5 |
|---|---|---|---|---|
| Light, 8 days | $2.24 | $4.48 | $11.20 | $22.40 |
| Moderate, 18 days | $39.96 | $79.92 | $199.80 | $399.60 |
| Heavy, 21 days | $384.72 | $769.44 | $1,923.60 | $3,847.20 |
This holds token usage constant to isolate price. It is not a benchmark of how well each model performs the same task: different models may take different numbers of turns and produce different outcomes. Use your actual distribution of usage for a budget, and investigate changes in the bill before treating them as waste.
What does a subscription seat change?
A subscription changes what the estimate means. Check Claude's current plan terms for the seat price, included usage, limits and extra-usage options. Tokens multiplied by API list rates are an estimate of equivalent API cost, not the charge for a subscription seat.
A high estimate alone does not prove a seat saved that amount or can cover the same workload. Limits, model availability and extra usage matter. Compare the actual invoice and delivered work as well as the estimate; a seat's value is not established by token volume.
How does this compare to the invoice?
The invoice includes your actual billing arrangement. These examples exclude discounts, taxes, other tool charges and subscription fees. Keep estimated and billed amounts distinct, and never add two views of the same usage together.
For a real team, collect usage per engineer, retain the model and token categories, and compare the same reporting window. Kibble (usekibble.com) is a Claude Code and Codex usage dashboard for teams: it applies this method to collected counts while keeping repository attribution an estimate.
Kibble runs this arithmetic on your team's real transcripts, per engineer, model and repo, every day, and merges matching Claude Console estimates without double-counting.
Start free- How much does GitHub Copilot cost per engineer per month?
- ccusage for a team: what it covers and where it stops
- Where Claude Code and Codex keep their usage logs, and what a token count is worth
- Estimated vs billed: why a coding-agent cost dashboard must never add the two
- Kibble vs an LLM gateway vs a local usage tool, in one table