What does Claude Code cost per day?
Paste your ccusage report or the Claude Code /usage block. The page prices the same tokens on Outrun and puts them beside the list-price figure your tools show. It is an estimate, worked out in your browser.
Paste your usage
Run npx ccusage@latest daily --json (or plain ccusage daily) and paste the output. You can also paste the Session block from /usage inside Claude Code.
Nothing you paste leaves this page. ccusage is a free, open-source tool that reads Claude Code’s local logs.
Prices as of , from the public price feed.
Outrun prices are exact feed rates; cache writes use the 5-minute cache-write rate. The list-price figure is the one ccusage or Claude Code reports, not a number we work out. Older model versions in your report (for example a Sonnet 4 model) are priced as the newest model of the same family on Outrun.
A sample Claude Code day, priced per model
An example day of agent work: 20M cached input tokens, 1.5M cache-write tokens, 300k fresh input tokens and 400k output tokens. Coding agents reread their history on every turn, so cached input is most of the volume. Rates are per 1M tokens; the cheapest route of each Claude model is shown.
| Model | Input | Output | Cache read | Cache write | Sample day |
|---|---|---|---|---|---|
Claude Opus 5.5 (2)claude-opus-5-5-2 | $0.80 | $4.00 | $0.04 | $1.00 | $4.14 |
Claude Sonnet 5.5 (2)claude-sonnet-5-5-2 | $0.40 | $2.00 | $0.02 | $0.50 | $2.07 |
Claude Haiku 5.5 (2)claude-haiku-5-5-2 | $0.02 | $0.10 | $0.002 | $0.025 | $0.123 |
Claude Fable 5.1 (2)claude-fable-5-1-2 | $2.00 | $10.00 | $0.05 | $2.50 | $9.35 |
DeepSeek V4.1 Flashdeepseek-v4.1-flash | $0.30 | $1.20 | $0.006 | $0.30 | $1.14 |
GLM-5.3glm-5.3 | $0.6552 | $2.0592 | $0.1216 | $0.6552 | $4.44 |
Kimi K3.1 Prokimi-k3.1-pro | $1.20 | $1.20 | $0.20 | $1.20 | $6.64 |
GPT-6.1 Solgpt-6.1-sol | $0.1176 | $0.588 | $0.00588 | $0.147 | $0.609 |
For scale: Anthropic’s Claude Code docs put the average at about $13 per developer per active day at list price, and under $30 per active day for 90% of users (Claude Code costs). The other models run in Claude Code through the same base URL; see using other models in Claude Code.
How the estimate works
Claude Code reports four kinds of tokens. Each one is priced at its own Outrun rate and added up.
ccusage JSON
dailyrows give tokens per day. WithmodelBreakdowns, each model is priced as its own family; days with any usage count as active days.ccusage table
The dated rows of the plain table are read the same way. The Total row is skipped so nothing counts twice.
/usage Session block
One session, priced per model line. Claude Code works out its own figure at list price; on a Pro or Max plan that figure is not your bill.
Rates
The cheapest route of each Claude model in the public price feed. Where a model has no cache rate, cache tokens are priced at its input rate.
Claude Code cost questions
Other workloads? Use the pricing calculator.
Why is Claude Code expensive at list price?
It resends the whole conversation, including files it has read, on every turn. Most of those tokens are cache reads, which are cheaper, but a long session still moves tens of millions of tokens a day.
Is my pasted usage sent anywhere?
No. The page reads your paste in your browser and only downloads the public price list. Nothing is uploaded.
How do I use Claude Code with Outrun?
Set ANTHROPIC_BASE_URL to https://api.outrun.to and ANTHROPIC_AUTH_TOKEN to your Outrun key. The setup post has the exact lines.
Which model should I use to save the most?
Sonnet handles most coding work for less than Opus, and Haiku suits small tasks and subagents. The table above shows the same day on each model.