Claude Code pricing in 2026: what it costs per developer
Claude Code has no price of its own. It comes with every paid Claude plan: Pro at $20 a month ($17 billed annually), Max at $100, Ultra at $200, and Team seats at $25 (Standard) or $125 (Premium) per user per month — all with usage limits. Or you pay per token through the API, where Claude Opus 5.5 is $4 per million input tokens and $20 per million output. On API billing, a typical developer on our own team used about $86 a month of model usage with routing, at list prices, and a heavy one about $261.
The official plans that include Claude Code
Anthropic's plans share one usage pool between the Claude apps and Claude Code. When you reach a limit you wait for it to reset, move up a plan, or turn on usage credits at standard API rates.
| Plan | Price | Usage |
|---|---|---|
| Claude Pro | $20/month billed monthly; $17/month billed annually ($200 up front) | More than Free; limits every 5 hours and every week |
| Claude Max | $100/month | 5× Pro usage every 5 hours and every week; $100/month in API credits |
| Claude Ultra | $200/month | 10× Pro weekly usage, no 5-hour limit for verified accounts; $200/month in API credits |
| Claude Team — Standard seat | $25/seat/month billed monthly; $20 billed annually (2–150 seats) | More usage than Pro |
| Claude Team — Premium seat | $125/seat/month billed monthly; $100 billed annually | 5× a Standard seat |
| Claude Enterprise | $20/seat/month billed annually, 20-seat minimum, plus all usage at API rates | Pay for what you use; admins set user and org spend limits |
| Claude API (pay per token) | Per million tokens, input / output: Opus 5.5 $4 / $20; Sonnet 5.5 $2 / $10; Haiku 5.5 $0.10 / $0.50 (prompts up to 100K tokens) | No 5-hour or weekly limits (API rate limits apply); cache reads are cheaper (Opus 5.5 $0.20, Sonnet 5.5 $0.10 per million) |
Other coding tools teams compare it with
| Plan | Price | How model usage is billed |
|---|---|---|
| Cursor Teams | From $30/user/month | Third-party models such as Claude at API list price, plus a Cursor Token Rate of $0.25 per million tokens on eligible third-party requests (also when you use your own key) |
| GitHub Copilot Business | $19/granted seat/month | 1,900 AI credits per user per month into a shared pool; beyond it $0.01 per credit |
| CodeRouter Team (ours) | $19/developer/month | All model usage at list price +8%, from a prepaid team balance; budgets per member, project and company |
Sources: claude.com/pricing, cursor.com/pricing, Cursor Token Rate, GitHub Copilot plans — checked 2026-11-03. Prices change often; the vendor pages are authoritative.
What a developer actually uses
Plan prices don't tell you what the work costs. Here is what our own engineering team used, measured with one API key per engineer from May to September 2026: 385 active developer-days with at least 20 requests. All of it went through CodeRouter, so "routed model cost" is what the models CodeRouter picked cost at their list prices. This is our own team's usage, not customer data — a small team building a router, using Claude Code and other agents every day.
| Per developer | Typical (median) | Heavy (75th percentile) |
|---|---|---|
| Requests per active day | 163 | 378 |
| Routed model cost per active day | $4.30 | $13.03 |
| Requests per month (20 working days) | ~3,300 | ~7,600 |
| Routed model cost per month | ~$86 | ~$261 |
Each request at its model's list price: uncached input, cache writes, cache reads and output. Claude cache writes weren't logged before November 2026, so for May–September they are estimated at 6.6% of the Claude prompt, the share we measured in November; without that estimate a typical day is $4.17 and a heavy day $11.52. 79% of the prompt tokens were cache reads. Over the whole period it came to $5,581 ($4,574 without the estimate). Medians and 75th percentiles are taken separately for each row; a month is 20 working days. It shows what was spent, not what got solved.
Corrected 2026-11-07. This page used to show $4.34 for a typical day and $17.01 for a heavy one. Those were CodeRouter's recorded costs, not list prices: until October 29, 2026 CodeRouter's model table priced Claude Opus 4.7 at 3× its list price. We also removed a comparison that priced the same requests at Opus 5.5 and Sonnet 5.5 prices (see below).
Why we don't price this traffic on a single model
It's tempting to take these tokens and price them at Opus or Sonnet list prices to see what an all-Opus or all-Sonnet setup would cost. That doesn't work: a different model takes a different number of steps to finish the same task, so the same requests wouldn't be sent. Comparing models takes the same tasks, run on each setup.
How a day's cost adds up
For model-against-model cost, see a controlled test. In TokenWatt's 176-task coding benchmark — TokenWatt is a measurement project from the same team as CodeRouter — routed requests solved 96–100% of tasks for 56–70% less than sending every request to Opus, across runs. Those are single-file tasks with short, single-turn contexts, not long agent sessions. The method, the model-by-model results and the limits are on the benchmark page.
Flat plan or API billing: which is the better deal?
For an individual who uses Claude Code heavily, a flat plan usually wins — as long as the work fits in its limits. A typical month from our team costs ~$86 with routing at list prices, a little under Max ($100). A heavy month costs ~$261, more than Ultra at $200 flat. Anthropic doesn't publish its limits as token amounts, so we can't tell you exactly how much fits; Max gives 5× Pro usage per 5-hour window and per week, and Ultra 10× Pro per week with no 5-hour limit for verified accounts.
CodeRouter's pay as you go doesn't beat that for a heavy individual either: there is no monthly fee and no included usage, and every request is billed at model list price +8%.
| Per developer per month | Typical developer | Heavy developer | Catch |
|---|---|---|---|
| Claude Max / Ultra | $100 / $200 | $100 / $200 | Usage limits; Claude models only |
| Claude Team Premium seat (monthly) | $125 | $125 | Usage limits per seat |
| API with routing, at list prices (our team) | ~$86 | ~$261 | No plan limits; any model; you pay for what you send |
| CodeRouter Team (seat + routed usage at list +8%) | ~$112 | ~$301 | Pays per token; budgets instead of plan limits |
CodeRouter Team = $19 + routed cost × 1.08: $19 + $86 × 1.08 ≈ $112; $19 + $261 × 1.08 ≈ $301. Using our team's usage above; yours will differ.
When API billing with routing makes sense
- Your company pays API rates anyway. Claude Enterprise is $20 a seat plus all usage at API rates; Cursor Teams bills Claude at list price plus its token rate; Console API keys are pay per token. Then the model mix decides the bill, and routing changes the mix.
- You use more than one tool or model family — Claude Code, Cursor, Aider, Cline — and want one bill and one set of keys.
- You need to see and cap spend per developer and per project, with alerts before a budget runs out.
- Work can't stop at a 5-hour or weekly limit. API billing has no plan limits; you set the budgets.
If none of those apply and you are one person on Claude models, take the flat plan that fits your usage. If they do, see CodeRouter for Teams or the pricing page.
FAQ
How much does Claude Code cost per month?
Claude Code has no price of its own: it is included in every paid Claude plan. As of 2026-11-03 that is Pro at $20 a month ($17 a month billed annually), Max at $100, Ultra at $200, and Team seats at $25 (Standard) or $125 (Premium) per user per month billed monthly — all with usage limits shared with the Claude apps. On API billing you pay per token instead: on CodeRouter's own team a typical developer used about $86 of model usage a month with routing, and a heavy one about $261, at list prices.
Is Claude Code cheaper with an API key or a subscription?
For one person who uses Claude Code heavily, a subscription is usually cheaper as long as the work fits in the plan's limits: a heavy developer on our team used about $261 a month at API list prices, against $100 for Max and $200 for Ultra. API billing makes more sense when your company pays API rates anyway (Claude Enterprise bills all usage at API rates), when you need budgets and spend per developer, when you use several tools or model families, or when work must not stop at a plan limit. On fixed tasks, routing lowered the bill: in TokenWatt's 176-task benchmark (TokenWatt is a measurement project from the same team as CodeRouter), routed requests solved 96–100% of tasks for 56–70% less than sending every request to Opus, across runs. Those are single-file tasks, not long agent sessions.
How do teams track Claude Code spend per developer?
Give every engineer their own API key and attribute spend per key. Claude Enterprise lets admins set user and organization spend limits. CodeRouter Team (our product) gives each engineer their own key, shows spend per member, project and model, and sets monthly budgets per member, per project and company-wide, with an email alert at 80% and a downgrade to cheaper models or a pause at 100%. Admins see spend and metadata, never prompts or responses.
How much does Claude Code cost per day on the API?
On CodeRouter's own team (May–September 2026, 385 active developer-days), the median developer-day cost $4.30 with routing and a heavy day (75th percentile) $13.03, at list prices. Your number depends on how many requests you send, how much context each one carries — an agent resends its context every turn, and 79% of our prompt tokens were cache reads — and which model answers each one.
What does Claude Code cost for a team of 10?
On seat plans: 10 Claude Team Standard seats are $250 a month billed monthly and 10 Premium seats $1,250, each with usage limits. On usage billing it depends on use: at our team's typical usage (about $86 of routed model cost per developer per month), CodeRouter Team would be about $1,119 a month (10 × $19 seats plus usage at list price +8%). Seat plans are cheaper when everyone stays within the limits; usage billing has no plan limits, shows spend per developer and lets you set the budgets.
How do I lower Claude Code's API bill?
Keep each session on one model so the prompt cache stays warm, send easy turns to cheaper models, and keep the frontier model for the turns that need it. A router does this per request: point Claude Code at CodeRouter with ANTHROPIC_BASE_URL=https://www.coderouter.io, ANTHROPIC_AUTH_TOKEN set to your cr_ key (not ANTHROPIC_API_KEY) and ANTHROPIC_MODEL=auto. On fixed tasks the effect is measured: in TokenWatt's 176-task benchmark, routed requests solved 96–100% of tasks for 56–70% less than sending every request to Opus (single-file tasks). In a long agent session, switching models mid-session throws the prompt cache away, so a router has to count that before it switches.
Sources
- Claude plans and API prices: https://claude.com/pricing (checked 2026-11-03)
- Cursor Teams: https://cursor.com/pricing and https://cursor.com/help/models-and-usage/token-rate (checked 2026-11-03)
- GitHub Copilot Business: https://docs.github.com/en/copilot/get-started/plans (checked 2026-11-03)
- Our team's usage and the benchmark aggregates: coderouter-176-2026-11.json (key
team_usage_2026_05_09) - CodeRouter prices: /pricing