Claude Code cost per developer
Verified Source Anthropic's Claude Code cost documentationPrices from the Claude API price list
On API billing, Claude Code costs Anthropic's enterprise users an average of about $13 per developer per active day, which the same source puts at $150 to $250 per developer per month. 90% of users stay under $30 per day. Subscription users (Pro, Max, Team, Enterprise seats) pay a flat seat price instead and hit rate limits rather than a bill.
Anthropic's published figures
| Figure | Value | Source |
|---|---|---|
| Average cost per developer per active day (enterprise API) | about $13 | Anthropic |
| Typical monthly cost per developer | $150 to $250 | Anthropic |
| Daily cost ceiling for 90% of users | under $30 | Anthropic |
| Background token usage (haiku summaries, command processing) | under $0.04 per session | Anthropic |
| Agent teams token multiplier versus a single session | about 7 times | Anthropic |
The token prices behind the bill
Claude Code defaults to Sonnet-class models for most work and can be switched to Opus. Every turn re-sends the conversation, so prompt caching does most of the work of keeping cost down.
| Model | Input per 1M | Output per 1M | Cache write 5m per 1M | Cache read per 1M |
|---|---|---|---|---|
| Claude Sonnet 5 | $2.00 | $10.00 | $2.50 | $0.20 |
| Claude Opus 5 | $5.00 | $25.00 | $6.25 | $0.50 |
Worked example: one working day
Assume 30 agentic turns in a day. Each turn reads a 40,000 token cached context (repository files, prior conversation), adds 3,000 tokens of fresh input, writes 2,000 new tokens to the cache and produces 1,500 output tokens.
| Model | input | output | cache read | cache write 5m | Per turn | Per day (30 turns) | Per month (20 working days) |
|---|---|---|---|---|---|---|---|
| Claude Sonnet 5 | $0.006000 | $0.0150 | $0.008000 | $0.005000 | $0.0340 | $1.02 | $20.40 |
| Claude Opus 5 | $0.0150 | $0.0375 | $0.0200 | $0.0125 | $0.0850 | $2.55 | $51.00 |
What moves the number
- Model choice. Opus 5 costs five times Sonnet 5 on every token dimension. Keep Opus for planning and review turns.
- Context size. Cost per turn grows with everything in context. Use
/compactor/clearbetween tasks; the docs recommend this specifically. - Cache hit rate. Cache reads are 10% of input price but expire after 5 minutes of inactivity. Long pauses mid-task re-pay the write.
- Agent teams and subagents. Parallel agents each carry their own context. Anthropic's figure of about seven times is the planning number.
- Spend controls. Console admins can set per-workspace spend limits; team plans can set per-seat extra usage caps.
Pricing a Claude Code rollout for a team? Multiply the per-day figure by active developer days, not headcount. Use the Claude calculator with your own context sizes, and put the result in a budget contract if the seats are API-billed.