How much Claude Code costs per month via API: three usage profiles

With Claude Code, pricing via API comes down not to a plan but to how many tokens the agent reads and writes. Below: the formula, three load profiles (a light, medium and heavy month) on Claude Opus 5.5 and Claude Sonnet 5, a comparison with Anthropic's subscription, and ways to cap spending. All Zukko prices are as of October 7, 2026.

What the bill is made of

On every request Claude Code sends the model the whole current context: the system prompt, contents of files it has read, conversation history and tool results. Most of this context repeats from request to request, so prompt caching kicks in. The bill has four lines:

• fresh input — the new part of the context that isn't cached;
• cache write — when a chunk of context is cached for the first time;
• cache read — re-sent context that is already in the cache;
• output — the model's answer, including reasoning tokens (per the Claude Code docs, thinking tokens are billed as output).

Formula: cost = fresh input × input price + cache write × write price + cache read × read price + output × output price, all volumes in millions of tokens.

Claude Code uses the Anthropic format (/v1/messages), so the gateway sees both cache reads and writes and bills them at their own prices. In the OpenAI format all Claude input would be billed at the input price — that doesn't apply to Claude Code, but it matters if you build your own agent.

Opus 5.5 and Sonnet 5 prices

Prices as of October 7, 2026, USD per 1M tokens:

The intelligence index is the Artificial Analysis Intelligence Index shown in the Zukko catalog. For the current list and prices, call GET https://api.zukko.pro/v1/models or open the Models page in the dashboard.

Claude Code pricing for a month: three profiles

Real usage depends on the project, so below are assumptions, not measurements. The proportions follow an example in the Claude Code docs: the sample /usage output there shows 940 thousand cache-read tokens, 50 thousand cache-write, 5.3 thousand output and 1.2 thousand fresh input. So cache reads are the bulk, writes are a few percent of that, and output and fresh input are smaller still. We padded output to allow for reasoning.

One medium day on Sonnet 5, line by line:

• cache read: 10 × 0.05 = $0.50;
• cache write: 0.6 × 0.5 = $0.30;
• fresh input: 0.05 × 0.4 = $0.02;
• output: 0.15 × 2 = $0.30;
• total $1.12 per day, $22.40 for 20 days.

The same day on Opus 5.5: 10 × 0.08 + 0.6 × 1 + 0.05 × 0.8 + 0.15 × 4 = 0.80 + 0.60 + 0.04 + 0.60 = $2.04, or $40.80 per month.

All profiles (an example calculation, prices as of October 7, 2026):

For comparison: if the whole medium-profile context were billed at the input price with no cache, Sonnet 5 would cost (0.05 + 0.6 + 10) × 0.4 + 0.15 × 2 = $4.56 per day and $91.20 per month. So what drives the bill most is how often the context is re-read and how often the cache misses.

The same volumes at Anthropic's official prices (claude.com/pricing as of October 7, 2026: Opus 5.5 — $4 / $20, cache read $0.20, write $5; Sonnet 5 — $2 / $10, read $0.20, write $2.50) come to $164 on Opus 5.5 and $102 on Sonnet 5 for the medium profile. Same formula, different prices.

How to find your own profile

Don't guess: work for a day and check /usage in Claude Code. Its Session block shows input, output, cache read and cache write per model. Plug them into the formula and multiply by your working days. The dollar figure Claude Code shows is computed at Anthropic list prices, so check the actual charges in the Zukko dashboard under Usage.

Comparison with a subscription

On claude.com/pricing as of October 7, 2026: Pro is $20 per month billed monthly, or $17 per month billed annually ($200 up front); Max starts at $100 per month, with a choice of 5x or 20x Pro usage. Claude Code is included in Pro and Max. Prices exclude tax, and Anthropic may change them.

The key difference is the billing model, not the amount:

• Subscription — a fixed fee plus usage limits. Per the Claude Code docs, when you run out you get a session-limit or weekly-limit message and work stops until the window resets (or extra usage credits are billed, if enabled).
• API — you pay for actual tokens; Zukko has no usage windows or weekly limits, only your balance and the limits you set on the key yourself.

A direct comparison isn't possible: subscription limits aren't expressed in dollars. A practical rule: if your real month by the formula comes out well below the subscription price (like the light profile), the API is cheaper; if you regularly hit subscription limits, the API removes that ceiling, but the bill grows with the load.

How to cap spending

Key limits

Create a separate key for Claude Code, set daily and monthly dollar limits, and keep only the models you need in its allowed list. When the cap is reached, the gateway returns 429 with key_daily_limit_reached or key_monthly_limit_reached. Limits are soft: the last request let through may slightly exceed the cap. Details in the key limits and cache docs↗.

Habits inside Claude Code

Per the Claude Code docs:

• /clear between unrelated tasks — old context stops being billed on every request;
• /compact shrinks history, but is itself a large request;
• /model switches models mid-session: Sonnet 5 for routine work, Opus 5.5 for hard tasks;
• /context shows what's taking up context;
• in claude -p mode, the --max-budget-usd flag stops work based on estimated spend.

Mind the cache: with an API key, the default cache lifetime is five minutes. After a break, the first request writes the context to the cache again at the write price.

Connecting

export ANTHROPIC_BASE_URL=https://api.zukko.pro
export ANTHROPIC_AUTH_TOKEN=fetch_your_key
export ANTHROPIC_MODEL=claude-sonnet-5
export ANTHROPIC_DEFAULT_HAIKU_MODEL=claude-sonnet-5
export ANTHROPIC_DEFAULT_SONNET_MODEL=claude-sonnet-5
export ANTHROPIC_DEFAULT_OPUS_MODEL=claude-opus-5-5
claude

The address has no /v1 — Claude Code adds it. The key goes into ANTHROPIC_AUTH_TOKEN. ANTHROPIC_DEFAULT_HAIKU_MODEL is required: Claude Code sends background requests to the Haiku-slot model (by default claude-haiku-*), which is not in the Zukko catalog, and the gateway answers an unknown model name with 404 model_not_found. The Sonnet and Opus slot variables make /model sonnet and /model opus point to catalog models. Per the Claude Code docs, if you set only the base URL without a credential, a saved claude.ai login stays active along with its limits. Step by step in the Claude Code guide↗; more savings tips in the article how to cut token usage in Claude Code↗.

To test your own profile in practice, create a key in the dashboard↗ with a small daily limit and work with it for a week. Payment from Russia is available: SBP, Russian cards, crypto.

FAQ

How much does Claude Code cost per month if I use it daily?

Under our assumptions, the medium profile (20 working days) is about $22.40 on Sonnet 5 and $40.80 on Opus 5.5 at prices as of October 7, 2026. Your number depends on project size and habits: take one day of /usage and plug it into the formula.

Why does caching affect the price so much?

Claude Code re-reads the whole context on every request. A cache read on Sonnet 5 costs $0.05 per 1M versus $0.4 for fresh input; on Opus 5.5, $0.08 versus $0.8. The larger the cache-read share, the lower the bill.

Opus 5.5 or Sonnet 5?

Sonnet 5 is half the price on every line. Opus 5.5 has an intelligence index of 58; there is no index data for Sonnet 5. Test both on a few of your own tasks and switch with /model as needed. How to run such a test on your own repository is covered in the article the best AI model for programming↗.

Can I block expensive models on the Claude Code key?

Yes. Set the list of allowed models in the key settings. A request to a model outside the list gets 403 model_not_allowed_for_key and isn't billed.