Coding Tools
Claude subscription or API: which costs less
Compare Claude Pro, Max and Ultra plan prices with per-token API prices, and find the cost crossover for your own workload
Short answer: a Claude subscription (Pro, Max, Ultra) pays for the Claude apps and Claude Code under usage limits; the Claude API is a separate product billed per token. If your own code calls the model — an agent, a pipeline, a feature in your product — you need the API. If you only work inside the Claude apps by hand, a subscription is usually the cheaper way to buy a lot of chat. The tables below convert each plan price into API tokens at today's prices, so you can compare against your own usage.
Two products, two bills
Anthropic bills the two separately: "A paid Claude subscription enhances your chat experience, and the Claude API is billed separately." Max, Ultra and Team plans include a monthly amount of API credit that you can spend in Console; Pro does not.
That split decides most cases before price does:
| You are doing | What you need |
|---|---|
| Chatting, writing, reviewing, or running Claude Code interactively | Either: a subscription plan, or an API key |
| Code that calls the model: agents, batch jobs, product features, evaluation runs | The API, with a key you control; Claude Code itself also accepts an API key |
| The same work with more than one vendor's models | An API gateway or the vendors' APIs |
| Sharing one budget across a team without per-seat plans | The API, with one key per project |
What each plan costs
Anthropic's list prices for individual plans (read from the pricing page on 2026-11-13):
| Plan | Price | What it adds |
|---|---|---|
| Free | $0 | Limited usage every 5 hours and every week |
| Pro | $20 per month billed monthly; $17 per month on the annual plan, $200 up front | More usage than Free, every 5 hours and every week; Claude Code |
| Max | $100 per month | 5x Pro usage; $100 a month in API credits |
| Ultra | $200 per month | 10x Pro usage, no 5-hour limit for verified accounts; $200 a month in API credits |
Usage limits apply on every plan, and how fast you reach one depends on the model, how long your chats are and how large your files are.
The monthly API credits come with three conditions worth knowing before you count them as free tokens:
- They are claimed by linking a Console organization to the plan and arrive each billing cycle.
- They expire at the end of the billing cycle, with no rollover.
- They cover the Claude API, Managed Agents, the Agent SDK and the playground, not Claude Code or extra usage in the Claude apps.
What a plan price buys in API tokens
The API charges per token: input tokens (what you send) and output tokens (what the model writes) are priced separately, and output is the expensive side. Prices below are for two current Claude models on TokenLab, at the Verified price and at the model maker's official price. Use the Models page for the current numbers; prices change.
| Model and route | Input / output per 1M tokens | Blended cost per 1M tokens* | Tokens for $20 | For $100 | For $200 |
|---|---|---|---|---|---|
| claude-sonnet-5-5, Verified | $0.70 / $3.50 | $1.26 | 15.9M | 79M | 159M |
| claude-sonnet-5-5, Official | $2.00 / $10.00 | $3.60 | 5.6M | 28M | 56M |
| claude-opus-5-5, Verified | $1.40 / $7.00 | $2.52 | 7.9M | 40M | 79M |
| claude-opus-5-5, Official | $4.00 / $20.00 | $7.20 | 2.8M | 14M | 28M |
* Four input tokens for every output token, a common ratio for coding and agent work. A different mix, prompt caching or long-context pricing moves the number; the formula is input × in-price + output × out-price.
The honest way to read this table: it is a cost reference, not an exchange rate. Anthropic does not publish a plan's allowance in tokens — the limits are described by time window, and the amount you get depends on the model and the size of your conversations. If you hit a weekly limit halfway through a task, that is the case where a token bill is worth paying for.
One coding-agent session, three ways to pay
A session that sends 150k input tokens and receives 15k output tokens:
| Route | Cost per session | Sessions per $20 | Per $200 |
|---|---|---|---|
| claude-sonnet-5-5, Verified | about $0.16 | about 127 | about 1,270 |
| claude-sonnet-5-5, Official | $0.45 | about 44 | about 444 |
| claude-opus-5-5, Official | $0.90 | about 22 | about 222 |
Long agent sessions resend history on every step, so the input side dominates and caching matters more than the per-token price you first looked at. See Control coding agent costs for the levers that move this number.
When the API wins on money
- Your volume is small: a few hundred thousand tokens a month costs less than any plan.
- Your work is automated: cron jobs, CI, batch classification, evaluation runs, anything that has to run without a person at the keyboard.
- You need several vendors: one key and one balance across Claude, GPT, Gemini and open models, so you can route each task to the cheapest model that finishes it.
- You need per-project limits: a separate key per project or tool, with its own cap, instead of one shared plan.
- You need exact accounting: per-request cost in Usage, not a plan price amortised by guesswork.
When the subscription wins
- You mostly chat and write in the Claude apps, and you value the flat monthly price.
- You use Claude Code interactively all day and rarely hit a limit.
- You want the plan's features (projects, file handling, desktop and mobile apps) rather than raw model access.
Both can be right at once: a plan for people, the API for the product.
Estimate your own bill in ten minutes
- Pick one representative task and run it five times on the same model, so the variance is visible.
- Open Usage in the workspace that owns the key and note input, output and cached tokens per run.
- Multiply by the number of runs you expect per month.
- Compare the result with the plan price in the table above, and remember that plan limits are time-windowed rather than token-counted.
Start with one key, no subscription
TokenLab has no subscription and no minimum spend: you top up a balance from $5, create a key, and pay only for the tokens each request uses. One key works across Claude, GPT, Gemini and open models, and Usage breaks the spend down by model and key. Read Billing and pricing for how charges, holds and the Verified and Official routes work.