Three ways to pay for Claude API access, compared
There are three ways to pay for Claude: Anthropic direct, a prepaid credit reseller, or a pay-as-you-go gateway. They are not the same product, and the lowest headline number is not always the one that costs you least, prepay, expiry and minimum top-up all change the answer.
| Route | Price | Prepay | Commitment |
|---|---|---|---|
| Anthropic direct | List price | None | None |
| Prepaid credit resellers | 20–25% of list | $250 – $10,000 | Buy a bundle |
| Pay-as-you-go gateway | 50% of Anthropic list (verified) | $1 minimum | None |
Claude prices compared
| Model | Official in / out | Ours in / out | Saving |
|---|---|---|---|
| Claude Opus 5 | 5 / 25 | 2.5 / 12.5 | -50% |
| Claude Sonnet 5 | 2 / 10 | 1 / 5 | -50% |
| Claude Fable 5 | 10 / 50 | 5 / 25 | -50% |
| Claude Haiku 4.5 | 1 / 5 | 0.5 / 2.5 | -50% |
| Claude Sonnet 4.6 | 3 / 15 | 1.5 / 7.5 | -50% |
| Claude Opus 4.8 | 5 / 25 | 2.5 / 12.5 | -50% |
USD per 1M tokens, verified 2026-11-02.
Which one should you pick
- Spend under $100 a month? Pay-as-you-go at 50% of list. Prepaying locks up capital for a discount you may not live long enough to use.
- Spend over $2,000 a month and can commit capital? Run the numbers on a prepaid bundle, the deeper discount may win.
- Need an enterprise contract, SLA or SOC 2? Stay with Anthropic direct. Neither reseller route replaces that.
What each route costs at your volume
| Monthly spend | Anthropic direct | Pay-as-you-go gateway | Worth switching? |
|---|---|---|---|
| $40 | $40 | $20 | Yes — no capital tied up |
| $400 | $400 | $200 | Yes |
| $2,000 | $2,000 | $1,000 | Yes, unless a prepaid bundle beats it after you price the capital |
Illustrative, at the 50% rate verified 2026-11-02. The gateway column assumes the current rate (re-check it before you commit, because upstream promotions move).
How to switch
Claude Code · Cursor · SDK
FAQ
Is a discounted Claude API route legitimate?
There is nothing inherently improper about routing your own requests through a reseller, you are still calling the same models. What matters is who holds the key and whether the rate is real. Check whether an official USD list price exists for the model and compare against it, which is what the table above does.
Why do some resellers advertise 75–80% off?
Those offers usually come from prepaid credit bundles You buy $250 to $10,000 of credit up front to earn the deeper discount. The saving is real but it is not free: you take on capital risk and counterparty risk, and you may never consume the credit.
What is the catch with pay-as-you-go?
The rate is higher than a deep prepaid bundle and it moves with upstream promotions. In exchange you keep your capital and can stop at any time. For most monthly spends under a few thousand dollars that trade is worth taking.
Does using a gateway change my integration?
No. It is a base URL change in the SDK or an environment variable in tools like Claude Code. Your request and response handling stays identical, see the setup guides linked above.
When should I stay with Anthropic direct?
When you need an enterprise agreement, an SLA, SOC 2 paperwork, or a contractual data-processing commitment. Neither reseller route replaces those, and the discount is not worth losing them.
How much does the Claude API cost per 1M tokens?
Through the pay-as-you-go route the Claude models here bill between $0.5 / $2.5 and $5 / $25 per 1M input/output tokens. Anthropic's own list price for the same models runs $1 / $5 upward. Input and output are billed separately on every model, and output carries the higher rate on all of them So the input number people quote is never the number that decides the bill. Rates checked 2026-11-02.
How much does Claude cost per month?
A moderate application month, 50M input and 10M output tokens, costs about $50.00 on Claude Haiku 4.5 through this route, against $100.00 at Anthropic list. Heavier agent use, 500M input and 100M output, moves it to about $500.00 versus $1,000.00 direct. Those are token costs only; there is no subscription on this route. Your own in/out split will differ — run it through the cost calculator rather than extrapolating from these figures.
Which Claude model has the lowest cost per token?
Claude Haiku 4.5 at $0.5 per 1M input and $2.5 per 1M output. But the lowest rate per token is not always the lowest cost per task: a model that needs fewer tokens to reach the same answer can come out cheaper overall, and output is priced above input on all of them, so a model that writes tersely wins more than its rate suggests. Test the top two on your own prompts — see how the ranking changes with the token mix.
Why is output more expensive than input on Claude?
Because generating a token costs more compute than reading one, and every model here prices the two directions separately. The multiple is not constant across the family either, which is why a workload that mostly writes (a coding agent, a summariser) lands on a different cost ranking than a workload that mostly reads. Price the two directions separately or you will pick the wrong model.