Claude Code Runs $200 a Month on the API. The $100 Plan CUTS It in Half 👇
Claude Code Runs $200 a Month on the API. The $100 Plan CUTS It in Half 👇 Anthropic changed Claude Code pricing twice in ten days. On October 30 it renamed Max 20x to Ultra. On November 9 it shipped Claude Fable 5.2 at
Claude Code Runs $200 a Month on the API. The $100 Plan CUTS It in Half 👇
Anthropic changed Claude Code pricing twice in ten days. On October 30 it renamed Max 20x to Ultra. On November 9 it shipped Claude Fable 5.2 at $8 input and $40 output per million tokens. Fable 5.1 cost $10 and $50.
Most pricing guides still show the old plan names. This guide uses the new ones. Every price below was checked on November 22, 2026.
The plans: $20 buys one unit of usage
Claude Code has no price of its own. It rides inside every paid Claude plan. Free gets none of it. Pro costs $20 a month, or $17 a month when billed yearly. Max costs $100 a month. Ultra runs $200 a month.
The usage scales in a straight line. Pro gives 1x. Max gives 5x. Ultra gives 10x of Pro usage each week.
Every step costs $20 per 1x. An upgrade buys headroom. The unit price stays the same.
Max and Ultra add two extras. Fable runs inside Claude Code up to half the weekly limit. And each plan bundles monthly API credits: $100 on Max, $200 on Ultra.
| Plan | Price per month | Weekly usage | Fable in Claude Code | API credits |
|---|---|---|---|---|
| Free | $0 | No Claude Code | No | None |
| Pro | $20, or $17 billed yearly | 1x | Usage credits only | None |
| Max | $100 | 5x Pro | Up to half the weekly limit | $100 |
| Ultra | $200 | 10x Pro | Up to half the weekly limit | $200 |
Ultra drops the 5-hour limit for verified accounts. That is the one real difference beyond size. Everything else is just more of the same allowance.
The API meter: what each token costs
Pay per token through an API key and the meter runs on four rates. The table lists them in dollars per million tokens. The rates come from Anthropic's pricing pages, re-read by an independent guide on November 16, 2026. Opus 5.5 is the default model in Claude Code.
| Model | Input | Cache read | Output |
|---|---|---|---|
| Fable 5.2 | $8 | $0.20 | $40 |
| Opus 5.5 | $4 | $0.20 | $20 |
| Sonnet 5.5 | $2 | $0.10 | $10 |
| Haiku 5.5 | $0.10 | $0.01 | $0.50 |
Cache reads are the quiet line. Claude Code resends the whole conversation on every request. The cached part bills at the read rate. On a long session, reads dwarf everything else. That fact drives every saving below.
The math: one daily developer, priced three ways
Take a daily developer on Opus 5.5. Two feature sessions plus three quick fixes a day. One independent calculator prices that day at $9.46. Over 21 working days that is about $199 a month.
Anthropic reports its own average across enterprise deployments. It is about $13 per developer per active day. The range is $150 to $250 a month, with 90 percent of users under $30 a day. At $13 a day, 21 working days cost $273 a month.
Now compare the flat plans. Max costs $100 a month. That is half the $199 the same work costs on the API. The catch is fit. Anthropic publishes usage as multiples, not token counts. Max covers the work only if it fits inside 5x Pro usage.
Pro costs $20 a month. It fits light use: a few short sessions a week, which one pricing guide puts at $10 to $30 of API value a month. Ultra costs $200 a month. It fits heavy daily work plus long agentic runs, and it drops the 5-hour limit for verified accounts.
Three ways to cut the bill
Clear the conversation. A 200-request Opus session grows to 921,000 tokens and costs $27.33. Cache reads are $18.73 of that, or 69 percent. Running /clear every 50 requests cuts the same session to $14.12. That is a 48 percent cut from one command.
Drop a model tier. Sonnet 5.5 costs half of Opus 5.5 at every session length. A 60-request feature session costs $3.99 on Opus and $1.99 on Sonnet. Stretch it to 200 requests and the gap stays: $27.33 on Opus, $13.66 on Sonnet. Use Opus for hard architecture and Sonnet for everyday coding.
Haiku 5.5 costs $0.10 input and $0.50 output. It is the cheapest model for mechanical edits. But prompts over 100,000 tokens bill at five times the base rate. A 60-request Haiku session crosses that line at request 21, and the API cache lasts five minutes, not one hour. Any break longer than five minutes rewrites the whole conversation.
What can go wrong
The monthly API credits look like free money. They are not. The $100 on Max and the $200 on Ultra cannot pay for Claude Code. They pay for API calls outside Claude Code. An API key in your shell also replaces your subscription login. Claude Code then bills API rates with no plan limit.
Check /status after you log in. Fast mode bills to usage credits on every plan, even when plan usage remains. Usage credits bought on or after December 7, 2026 expire 12 months after purchase.
Two more limits shape the choice. The 5-hour window starts with your first message and refills in full five hours later. The weekly limit resets at a fixed time per account. Both windows are shared with Claude chat on web and desktop. Heavy chat use eats the same pool as Claude Code.
Verdict
Light users should start on Pro at $20 a month. Daily developers should start on Max at $100. Ultra fits only heavy daily work where the 5-hour window blocks real tasks. Pay per token only when the month is light or spiky.
First step: type /usage in Claude Code today. Read the 5-hour bar and the weekly bar. The bars tell you which plan fits before the next bill does.
Follow along
I test AI tools and build in public:
- GitHub: https://github.com/ovu-lab
- YouTube: https://youtube.com/@oVuCyber
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.