Claude Code Pricing 2026: Plans, API Rates, and What It Costs Per Developer

Quick answer. Claude Code is included in every paid Claude plan — Pro at $17/month billed annually ($20 monthly), Max from $100/month, and Team seats at $20 or $100 per seat. There is no separate Claude Code subscription and no free tier. If you pay per API token instead, Anthropic's own enterprise data puts the real figure at about $13 per developer per active day, or $150–250 per developer per month.

One thing the plan table doesn't show: the top model tier can bill outside your plan. Depending on your plan and seat tier, Fable usage may draw on usage credits rather than your included limits — and in scripted, non-interactive runs it does so without asking.

Every price on this page was read from Anthropic's official pricing pages on 2026-09-26 and is linked to its source. Two of them had changed in the previous month, which is the reason we date them.

Claude Code pricing: the subscription plans

Source: claude.com/pricing, read 2026-09-26.

Three things follow from that table that are easy to miss.

There is no Claude Code–only plan. You are buying a Claude subscription that happens to include the terminal tool. If you only want Claude Code and never touch the chat interface, you still pay the full seat price.

Claude Code draws from the same usage pool as everything else. Anthropic's own FAQ puts it plainly: "Claude Code is included in all paid plans. It shares the same usage limits as the rest of your plan, so your work in the terminal and your chats draw from one pool." A heavy afternoon in the terminal reduces what you can do in the browser that evening.

Claude Code limits reset on a rolling window, not a monthly one. Paid plans use a rolling five-hour session window plus a weekly cap. That shape matters more than the headline price for anyone who works in bursts: you can exhaust a five-hour window before lunch and be blocked until it rolls, regardless of how much of the month is left.

Fable: the tier that can bill outside your plan

Claude Code's /model picker offers haiku, sonnet, opus, and fable. The first three draw on your plan's included allowance. Fable sometimes does too — and sometimes doesn't.

Anthropic's wording is deliberately conditional, and it is worth reading exactly as written rather than as a rule:

"Depending on your plan and seat tier, Fable usage can bill to usage credits instead of drawing on your plan's included limits. When it does, the /model picker shows 'Requires usage credits' on the Fable row."

— Model configuration, read 2026-09-26

So there is no flat plan-by-plan rule to quote, and you should be suspicious of any article that gives you one. The authoritative answer for your own account is in the picker: open /model and look at the Fable row. If it says "Requires usage credits," Fable bills outside your allowance. If it doesn't, it's included. That check takes two seconds and beats any table.

What is stated unconditionally is the consent behaviour, and this is the part worth acting on:

• Interactive sessions prompt you. Claude Code shows a consent prompt before a Fable request bills usage credits.
• Enterprise members on org billing don't see the prompt.
• Scripted runs never see it. In non-interactive mode (the -p flag) and through the Agent SDK, "Claude Code never shows the consent prompt. When a Fable request there would bill to usage credits, Claude Code bills it without asking."

That last one is the practical risk. An interactive user gets a decision point; a CI job or scheduled task with --model fable does not. If you are scripting Claude Code, check the picker on an interactive session first so you know which side of the line your account is on before the script runs unattended.

The rate gap is the reason any of this matters. Fable 5.1 is $10 input / $50 output per million tokens against Opus 5.5's $4 / $20 — two and a half times the output rate. Its cache reads are the cheapest ratio available (0.025× base input, $0.25/MTok), which softens long sessions with a warm cache, but nothing offsets output.

You also cannot economise by turning thinking off: thinking cannot be disabled on Fable models or on Opus 5.5. The session toggle, alwaysThinkingEnabled, and MAX_THINKING_TOKENS=0 all have no effect there. Effort levels (low through max) are the only control.

One more wrinkle: Fable runs safety classifiers that can decline a request and fall back to another model — biology-flagged requests to Opus 5, cybersecurity-flagged ones to Opus 4.8. The fallback attempt bills at the fallback model's rates.

When it is worth it: Anthropic positions Fable for "tasks larger than a single sitting" — long autonomous sessions, root-cause investigations, architecture decisions, and large codebases where its native 1M context matters. Fable is never the default on any plan; you select it deliberately with /model fable. For routine coding, Sonnet at $2/$10 does the job at a fifth of Opus 5.5's output cost and a twentieth of Fable's.

What it actually costs per developer

This is the number most pricing pages leave out, and it comes from Anthropic's own documentation rather than an estimate:

Across enterprise deployments, the average cost is around $13 per developer per active day and $150–250 per developer per month, with costs remaining below $30 per active day for 90% of users.

Source: Manage costs effectively, read 2026-09-26.

Put that next to the subscription table and the decision gets simpler than most comparisons make it. A Pro seat at $17–20/month is far below the $150–250 band, because a subscription seat caps your usage rather than metering it. The per-token route only makes sense when you need to exceed what a seat allows, or when you need per-user billing and spend controls that subscriptions don't provide.

The 90th-percentile figure is the one to plan around. An average of $13/day with a $30/day ceiling for 9 in 10 users tells you the distribution has a long tail — budgeting on the average will under-fund your heaviest users.

Paying by API token instead

If you authenticate Claude Code with an API key rather than a subscription, you pay per token at standard Claude API rates. As of 2026-09-26:

Source: platform.claude.com pricing, read 2026-09-26. MTok = one million tokens.

Two recent changes are worth flagging because stale guides still carry the old figures:

• Sonnet 5's $2/$10 rate is now permanent. It launched as introductory pricing set to expire on 2026-08-31, and the scheduled increase to $3/$15 was cancelled. Any article still quoting $3/$15 for Sonnet 5 predates that.
• Opus 5.5 is cheaper than the Opus 5 it succeeds — $4/$20 against $5/$25 — and its cache reads are half the usual rate (0.05× instead of 0.1×). Newer is not more expensive here.

Prompt caching is where the real arithmetic lives. Writing to the 5-minute cache costs 1.25× the base input rate and the 1-hour cache costs 2×, but reads cost a tenth of base input. A 5-minute cache pays for itself after a single read; the 1-hour cache needs two. In a Claude Code session, where the same conversation is re-sent on every request, cache reads are most of the input bill.

Batch API requests are 50% off both input and output, though that path does not apply to interactive terminal work.

Is there a Claude Code free tier?

No. The Free plan does not include Claude Code, and there is no separate free tier for the terminal tool. The nearest thing to free access is API credits: Anthropic states that "new users receive a small amount of free credits to test the API" — enough to try it, not enough to work with.

That makes the practical entry price $17/month (Pro, billed annually), which is the cheapest legitimate way to run Claude Code.

The costs that aren't on the pricing page

If you are paying per token, these are the eight things that move the bill without appearing in the rate table. All are documented; none are prominent.

1. The tokenizer changed, and the same text now costs ~30% more. Claude Opus 4.7 and later use a newer tokenizer that "produces approximately 30% more tokens for the same text." Per-token prices did not change to compensate, so a prompt that cost $1.00 on an older model can cost around $1.30 on a current one at the same nominal rate. Any cost baseline you measured before that change is wrong.

2. Long sessions re-send the whole conversation. Claude Code sends the full conversation with every request, and every tool call is another request carrying the entire history. A one-line question at 5pm in a session opened at 9am draws usage for the whole day's context — at cached rates, but not free.

3. Cache misses after a break reprocess everything. The prompt cache lifetime is one hour on a subscription, and it drops to five minutes once you are drawing on usage credits. On an API key it is five minutes by default. Your first message after a longer break misses the cache and reprocesses the full context at full input price.

4. Agent teams use roughly 7× the tokens. Each teammate runs its own Claude Code instance with its own context window. Anthropic's documentation puts the multiplier at "approximately 7x more tokens than standard sessions when teammates run in plan mode."

5. Scheduled tasks bill while you're idle. A scheduled task fires on its interval whether or not you are at the keyboard, and each firing sends your full context.

6. US-only inference costs 1.1×. Setting inference_geo: "us" for data residency applies a 1.1× multiplier to every token category — input, output, cache writes, and cache reads alike.

7. Fast mode roughly doubles the rate. Opus 5.5 in fast mode is $8/$40 against $20 standard for output; Opus 5 and 4.8 are $10/$50. It is a research preview on the first-party API only.

8. Regional endpoints on Bedrock and Google Cloud carry a 10% premium over global endpoints, for models from Sonnet 4.5 onward.

9. Fable can bill to usage credits, and scripted runs are never asked. Whether it does depends on your plan and seat tier — check the /model picker. Covered in full above.

Two more line items apply only if you use the matching tools: web search is $10 per 1,000 searches, and code execution is free alongside web search or web fetch but otherwise bills at $0.05 per container-hour after 1,550 free hours per organization per month. Web fetch adds no charge beyond the tokens it pulls in.

How to cut the bill

The documented levers, in rough order of effect:

Clear between unrelated tasks. /clear starts a fresh context. Carrying an old conversation into new work means paying for it on every request. Anthropic names long uncleared sessions and leaving Opus as the default model as the two usual causes of unexpectedly high spend.

Match the model to the job. Sonnet handles most coding work at a fifth of Opus 5.5's output rate. Reserve the Opus tier for architectural decisions and multi-step reasoning.

Lower the effort level on simple tasks. Thinking tokens bill as output tokens, and the default budget can run to tens of thousands per request. Use /effort for work that does not need deep reasoning — though note that thinking cannot be turned off on Opus 5.5 or the Fable models.

Move instructions out of CLAUDE.md into skills. CLAUDE.md loads into context at session start and stays there for work it has nothing to do with. Skills load only when invoked. Anthropic suggests keeping CLAUDE.md under 200 lines.

Delegate verbose operations to subagents. Test output and log processing stay in the subagent's context; only the summary returns to your main conversation.

Prefer CLI tools to MCP servers where both exist. gh, aws, and gcloud add no per-tool listing to context.

Run /usage to see where your tokens are going. On a paid plan it breaks usage down by skills, subagents, plugins, and individual MCP servers, and flags any behavior — long context, cache misses — accounting for 10% or more of recent usage.

FAQ

How much does Claude Code cost?

Claude Code is included in every paid Claude plan at no extra charge: $17/month for Pro billed annually ($20 monthly), from $100/month for Max, and $20 or $100 per seat on Team. If you pay per API token instead, Anthropic's enterprise data puts the real figure at about $13 per developer per active day, or $150–250 per developer per month.

Is Claude Code free?

No. The Free plan does not include Claude Code and there is no separate free tier. New API accounts get a small amount of free credits for testing. The cheapest real access is a Pro subscription at $17/month billed annually.

What are the Claude Code usage limits?

Paid plans use a rolling five-hour session window plus a weekly cap, and the allowance is shared across Claude Code, Claude chat, and Cowork — one pool, not three. The exact number of messages varies with conversation complexity and which model you use. Max plans offer 5× or 20× Pro's allowance. If you hit the limit, usage credits let you continue past it at metered rates.

Is Claude Code cheaper on a subscription or by API key?

A subscription, for almost everyone. A Pro seat is $17–20/month against a documented $150–250/month average for per-token usage — because a seat caps your usage while an API key meters it. The API route makes sense when you need to exceed what a seat allows, or when you need per-user spend reporting and workspace-level spend limits, which subscriptions don't provide.

Does Fable cost extra on top of my Claude Code subscription?

It depends on your plan and seat tier, and there is no universal rule — Anthropic's wording is that Fable usage "can bill to usage credits instead of drawing on your plan's included limits." Check the /model picker on your own account: when Fable bills outside your allowance, the Fable row is labelled "Requires usage credits." When it isn't labelled, it's included. Don't trust a plan-by-plan table for this, including ones you find in articles.

The part that holds regardless of plan: interactive sessions prompt for consent before billing credits, Enterprise members on org billing don't see that prompt, and non-interactive runs (the -p flag) and the Agent SDK never show it — they bill without asking. Check the picker before you put --model fable in a script.

How much more expensive is Fable than Opus?

Fable 5.1 is $10 input / $50 output per million tokens against Opus 5.5's $4 / $20 — two and a half times the output rate. Its cache reads are the cheapest available at $0.25/MTok (0.025× base input) versus Opus 5.5's $0.20 at 0.05×, so a long session with a warm cache narrows the gap on input, but not on output. You also cannot reduce cost by disabling thinking: that is not possible on Fable models or on Opus 5.5, and effort levels are the only control.

Which Claude model is cheapest for coding?

Of the current models, Claude Haiku 4.5 at $1/$5 per million tokens, then Claude Sonnet 5 at $2/$10. Sonnet is the practical choice for most coding work — Anthropic's own guidance is to reserve the Opus tier for complex architectural decisions and multi-step reasoning. Note that Opus 5.5 at $4/$20 is cheaper than the Opus 5 it replaced.

Why is my Claude Code usage higher than I expected?

Usually one of three things. A session left open for hours re-sends its whole conversation with every request. A first message after a break longer than the cache lifetime — one hour on a subscription, five minutes on usage credits or an API key — reprocesses the full context at full price. Or Opus is still set as the default model. Agent teams and scheduled tasks are the next most common causes; on a paid plan, /usage flags whichever behavior accounts for 10% or more of your recent usage.

Does Claude Code count against my Claude chat usage?

Yes. They draw from one pool. Anthropic's FAQ states that Claude Code shares the same usage limits as the rest of your plan, so terminal work and browser chats reduce the same allowance.

How were these prices verified?

Every figure was read on 2026-09-26 from four Anthropic sources: claude.com/pricing for subscription plans, platform.claude.com pricing for per-token API rates and feature pricing, code.claude.com/docs/en/costs for the per-developer cost figures and the usage-reduction guidance, and code.claude.com/docs/en/model-config for model availability by plan and the Fable usage-credit rules. Prices change without notice — two of the figures here had changed within the previous month — so confirm against those pages before committing a budget.