Claude Code pricing: plans, token rates and what uses credits

Claude Code plans, limits and per-token rates, plus a measured evaluation and one of our Kiro workflows showing which steps use the credits.

Pay by plan or by the token

Claude Code comes with the Pro, Max and Ultra plans, or you can pay per token with an API key. Plans fix the monthly price and the usage limits; per-token cost grows with every token unless you set a spend limit. Source: Anthropic pricing, checked 9 November 2026.

	Pro $20/mo	Max $100/mo	Ultra $200/mo	API Per token
Claude Code included	✓	✓	✓	✓
Usage every 5 hours	Pro limit	5x Pro	No 5-hour limit (verified accounts)	No plan limit
Usage every week	Pro limit	5x Pro	10x Pro	No plan limit
Monthly API credits	--	$100	$200	--
Annual billing	$17/mo ($200 up front)	--	--	--

When a plan beats paying per token

Anthropic puts the enterprise average at about $13 per developer per active day (Claude Code costs docs, checked 9 November 2026). At that rate, Pro equals about 1.5 active days a month and Ultra about 15, though plans carry 5-hour and weekly usage limits that paying per token does not.

Monthly cost: plan price vs pay per token at Anthropic's enterprise average ($13 per active day)

Current Claude model rates, per million tokens

Model	Input	Output	Batch in	Batch out	Cache read
Claude Fable 5	$10.00	$50.00	$5.00	$25.00	$1.00
Claude Fable 5.1	$10.00	$50.00	$5.00	$25.00	$0.25
Claude Haiku 4.5	$1.00	$5.00	$0.50	$2.50	$0.10
Claude Haiku 5.5	$0.10	$0.50	$0.05	$0.25	$0.01
Claude Opus 4.5	$5.00	$25.00	$2.50	$12.50	$0.50
Claude Opus 4.6	$5.00	$25.00	$2.50	$12.50	$0.50
Claude Opus 4.7	$5.00	$25.00	$2.50	$12.50	$0.50
Claude Opus 4.8	$5.00	$25.00	$2.50	$12.50	$0.50
Claude Opus 5	$5.00	$25.00	$2.50	$12.50	$0.50
Claude Opus 5.5	$4.00	$20.00	$2.00	$10.00	$0.20
Claude Sonnet 4.5	$3.00	$15.00	$1.50	$7.50	$0.30
Claude Sonnet 4.6	$3.00	$15.00	$1.50	$7.50	$0.30
Claude Sonnet 5	$2.00	$10.00	$1.00	$5.00	$0.20
Claude Sonnet 5.5	$2.00	$10.00	$1.00	$5.00	$0.10

All prices in USD per 1M tokens. Batch runs asynchronously at a reduced rate; cache read applies to reused input tokens. A dash means the provider does not publish that rate.

These are the live per-token rates for Claude models. Full table with batch and cache rates: Claude API pricing.

Best model for Claude Code: on our task, Haiku 5.5 passed for a twentieth of Sonnet's cost

The current Claude models did the same small coding task, each answer scored against a known correct answer with a pass mark of 80 out of 100. All three passed: Haiku 5.5 scored 95 at about a twentieth of Sonnet 5.5’s cost. Opus vs Sonnet: Opus 5.5 scored 100 against Sonnet’s 90, at about twice the cost.

Cost per 1,000 answers, same coding task, current models (evaluation 643504cd, 9 Nov 2026)

One model generation changed the answer: Haiku 4.5 failed, Haiku 5.5 passed

Output tokens are the pieces of text a model writes back, and each one is paid for. On the 4.5 models, Haiku ignored “reply with only the code” and wrote 505 output tokens, failing the pass mark; on the 5.5 models all three wrote the same 89-token function. Re-check your model choice when a new version ships.

Output tokens per answer, same coding task: 4.5 models (4 Nov) vs 5.5 models (9 Nov)

Let your agent pick the Claude model for each task

From inside Claude Code, your agent can score Haiku, Sonnet and Opus on your own task and read back each one’s cost and score. Figures are from evaluation 643504cd (9 November 2026; the 4.5 run is 742a9840); the prompt is an example and the summary line is ours.

Agent
engineer: which Claude model should I use for this coding task? (example prompt)
> run_evaluation models: claude-haiku-5-5, claude-sonnet-5-5, claude-opus-5-5 bar: 80
polling ... running ... done
claude-haiku-5-5: score 95 pass $0.0000562 89 output tokens
claude-sonnet-5-5: score 90 pass $0.001124 89 output tokens
claude-opus-5-5: score 100 pass $0.002248 89 output tokens
summary (written by us): all three pass; haiku-5-5 passes for about a twentieth of sonnet's cost

Long, busy sessions are what push a Claude Code bill up

Anthropic’s docs say Claude Code token costs scale with how much context Claude processes. The same pattern shows in a Kiro workflow of ours: a step with 44 turns (model decisions) and 111 tool calls (actions such as reads) cost 12.05 credits, a step with 7 turns cost 1.07.

Credits, turns and tool calls per step, one Kiro workflow of ours (agent analysis, 4 Nov 2026)

See which model your task actually needs

Score your own task across models and find the cheapest one that clears your pass mark, from your agent or the dashboard. On Pro and above.

See Pro

See where your tokens went with /usage

In Claude Code, /usage shows a session’s tokens and estimated cost, and on paid plans what share came from each subagent, plugin and MCP server, with suggested changes. The panel shows the format; its values are illustrative. Source: Claude Code costs docs, checked 9 November 2026.

Agent
/usage (illustrative format; values are examples)
Session: input tokens 412k, output tokens 38k, est. cost $1.84 (list price)
Plan usage: 5-hour window 42%, weekly 18%
Attribution: subagents 31%, MCP servers 12%, skills 6%
Suggested: 25% Trim CLAUDE.md, 49k tokens per request

Using Claude in Cursor

Cursor is now sold to new customers inside The X Subscription. Claude and other third-party models are billed at API list price from prepaid credit on Premium ($30/mo) and above. Sources: Cursor models and pricing and The X Subscription, checked 9 November 2026.

Cursor plans that can use Claude (billed at API list price from prepaid credit). Source: cursor.com, 9 Nov 2026

Spot the helper agent that holds up the rest

When Claude Code hands work to subagents (helper agents), an agent analysis shows if one is doing far more than the others. In a Kiro workflow of ours, splitting work by count left the slowest of three workers about 100.7 seconds behind the average.

Three parallel workers, one Kiro workflow of ours (agent analysis, 4 Nov 2026)

Ask where a Claude Code session's spend went

An agent analysis reads Claude Code sessions as well as Kiro workflows, and returns each finding with its cost or time and its cause. The findings shown are real, from the same Kiro workflow of ours; the prompt is an example and the tool calls are shortened.

Agent
engineer: what made this session expensive? (example prompt; findings from a Kiro workflow of ours)
> check_trace_eligibility, upload from disk, analyze_graph_start, analyze_graph (condensed)
money: worker 1 needed 44 turns where a matching worker needed 23
money: every step re-read the same documents, about 1.07 credits in all
time: one worker had far more items than the others, about 100.7 s of waiting
root cause of the wait: the step that divided the items by count

On Pro: for Claude Code users

Score Claude models on your own task
Haiku, Sonnet and Opus compared on cost, score and speed
Analyse Claude Code sessions step by step
Re-check your model choice on a schedule as Claude models change

Use it from Claude Code, over MCP (the protocol agents use to call outside tools)

Run checks from inside Claude Code
Add it to Claude Code as an MCP server
Analyse your Claude Code sessions for wasted turns and tokens

$129/mo $10 for your first week

Then $129/mo. Cancel anytime.

Plan prices and limits are from Anthropic's pages, checked 9 November 2026. Per-token rates come from our live model registry. Measured results are from one evaluation and one workflow of ours; your own task will score differently.

Questions

Is Claude Code free?

No. It is included with the Pro, Max and Ultra plans, not the Free plan. You can also use it with an API key and pay per token.

What is the Claude Code usage limit?

On Pro, usage is limited every 5 hours and every week. Max gives 5x Pro usage in both windows; Ultra has no 5-hour limit for verified accounts and 10x Pro usage each week. How fast you reach a limit depends on the model, how long your sessions are and how large your files are (Anthropic pricing, checked 9 November 2026).

Is Claude Code cheaper than Kiro?

That turns on your usage, since Claude Code caps use per window and Kiro gives a monthly pool of credits; our Kiro vs Claude Code comparison lists each plan from both tools in one table.

Plan or API: which is cheaper?

At Anthropic’s enterprise average of about $13 per developer per active day (Claude Code costs docs, checked 9 November 2026), Pro equals about 1.5 active days and Ultra about 15. Plans also carry 5-hour and weekly usage limits that paying per token does not.

What is Claude Code's default model, and how do I change it?

Anthropic’s docs name two common causes of unexpectedly high spend: long sessions that were never cleared, and Opus left as the default model (Claude Code costs docs, checked 9 November 2026).

Analyse your agent workflows from your agent over MCP (the protocol agents use to call outside tools), on Pro and above: see how agent evaluation works.