Claude Code Pricing in 2026: Plans, Usage Limits and API Costs

Claude Code has no price tag of its own. It rides on a Claude subscription, starting at Pro for $20/mo, and there is no free tier that unlocks it. Heavier terminal use pushes people up to Max, and a Team seat, priced per member, adds SSO and admin controls on top.

These figures were read on Anthropic's pricing page and the Claude Code product page on September 25, 2026. Claude Code is Anthropic's terminal coding agent, and the confusing part is not the plan grid, it is that chat and code draw from one shared usage pool, so a long debugging session can eat into the allowance you meant to spend talking to Claude later that day.

Toolradar data: the AI coding ranking evaluated 153 tools, updated September 2026. The setup guide on this site is how to use Claude Code, the product writeup is the Claude Code review, and the directory record is Claude Code on Toolradar.

Plan	Price	Best for	What changes the bill
Free	$0	Chat only, no Claude Code	Does not include Claude Code at all
Pro	$20/mo, or $17/mo annual	Short sessions on a small codebase	Shares one pool with chat, weekly cap
Max 5x	$100/mo	Daily terminal use	5x the Pro allowance, still poolwide
Max 20x	$200/mo	All-day agent work	20x the Pro allowance
Team Standard	$25/user/mo, or $20/user annual	A team that wants seats, not API keys	1.25x Pro per seat, 2-seat minimum
Team Premium	$125/user/mo, or $100/user annual	Heavy users on the same team	6.25x Pro per seat
Enterprise	From $20/seat/mo annual, plus usage	Governance, SSO, audit logs	Newer seat plans meter usage like the API
Console (API)	Pay per token	Building on Claude, not chatting	Priced per model, per million tokens

There is no free way into Claude Code

The Free plan covers chat on the web, desktop and mobile, web search, file creation, code execution and up to five projects, and it costs nothing. It does not include Claude Code. Anthropic's own product page draws the line plainly: Claude Code is included starting on Pro, not on Free.

That is the opposite of how Cursor or GitHub Copilot onboard people. Cursor's Hobby tier and Copilot's Free tier both let a developer try the agent for nothing before paying. With Claude Code, the cheapest way to open a terminal session is a paid subscription, so there is no zero-cost trial of the actual product, only of the chat app around it.

Pro shares its pool with chat

Pro is billed monthly, or $17/mo if the year is paid upfront. That price is not new: it is the same tier Anthropic has sold for chat since before Claude Code existed, now extended to cover the terminal agent too.

The one thing to understand before buying Pro for coding is the shared pool. Claude Code, Claude chat, Claude Design and Claude Science all draw from the same rolling five-hour session limit and the same weekly cap. Anthropic does not publish an exact prompt count for Pro, because usage is metered by tokens processed rather than messages sent, and a large file read burns the allowance faster than a short question does. A person who chats with Claude during the day and then opens Claude Code at night is spending from one shared meter, not two separate ones.

Pro suits short coding sprints in a small repo, the phrasing Anthropic itself uses on the product page. A single bug fix, a small script, or an occasional refactor fits inside that allowance most weeks. A daily multi-file agent habit usually does not, and that gap is the signal to look at Max.

Max 5x and Max 20x are multiples, not new features

Max comes in two sizes, and both buy the same product as Pro, just more of it. Max 5x is $100/mo and gives five times the Pro plan's per-session allowance. Max 20x is $200/mo and gives twenty times that allowance. Neither tier unlocks a capability Pro lacks; the terminal agent, the models and the tool access stay identical across all three.

The step from Pro to Max 5x is a five-times price jump for a five-times usage jump, so the per-token value does not change, only the size of the tank. Anthropic recommends the smaller Max tier for people who work with Claude across a variety of tasks most days, and the larger one for daily users who lean on Claude for most of their work, including long agent runs that would otherwise hit the Pro ceiling by lunchtime.

Skip Max until you can name the week Pro ran out, not the week you assume it will. Claude Code does not warn you in advance how close you are to a limit, so the honest test is whether you have actually been interrupted mid-task, not whether a heavier plan sounds safer.

Team seats add SSO, still metered per person

Team Standard is $25 per member per month billed monthly, cheaper on the annual rate shown above, with a two-seat minimum and room to grow to 150 seats before Enterprise becomes the only option. Team Premium is $125 per member per month monthly, or $100 annually, for members with a heavier workload.

Each seat carries its own allowance rather than a shared team pool: Standard gives 1.25x the Pro plan's per-session usage, and Premium gives 6.25x. A quiet teammate's leftover allowance does not top up a busy one's, so a five-person team where two live in Claude Code all day and three barely touch it needs a mix of seat tiers, not five identical Standard seats. Team plans add SSO, an admin console, enterprise search and centralized billing on top of the usage, none of which personal Pro or Max accounts include.

Claude Code ships on every Team seat, so a team does not pay extra to turn it on. The seat price is the whole story until someone wants more than the seat's included allowance, which is where usage credits come in.

Enterprise now has two different shapes

Enterprise pricing on claude.com/pricing lists a base seat rate billed annually, with usage on top metered at API rates. That is a newer, seat-based Enterprise plan, and it reads almost like a cheap Team seat with usage layered over it rather than folded into the price.

Anthropic's support documentation describes a second, older shape too: a usage-based Enterprise plan, including self-serve Enterprise, with no per-seat usage limits at all, because every token is billed at API rates from the start with no included allowance to run out of. Which shape an organization is on changes the whole budgeting conversation. A seat-based Enterprise customer watches a per-seat ceiling the way a Team customer does, while a usage-based Enterprise customer watches a token meter the way an API customer does. Neither shape has a single public number that applies to every buyer, so Anthropic's guidance for both is the same: talk to an account manager.

The API is a separate meter, priced per model

Outside a subscription, Claude Code can run against the Claude API through the Claude Console, and that path bills by the token, not by the seat. Prices below are per million tokens, read on Anthropic's API pricing docs on September 25, 2026.

Model	Input	Output	5-min cache write	Cache read
Claude Sonnet 5	$2/MTok	$10/MTok	$2.50/MTok	$0.20/MTok
Claude Opus 5.5	$4/MTok	$20/MTok	$5/MTok	$0.20/MTok
Claude Opus 5.5, fast mode	$8/MTok	$40/MTok	Stacks on top of fast pricing	Stacks on top of fast pricing
Claude Haiku 4.5	$1/MTok	$5/MTok	$1.25/MTok	$0.10/MTok

Sonnet 5 is the model Anthropic's own docs point developers toward for most coding work, with Opus reserved for harder architectural decisions. The API has no monthly ceiling and no shared pool with chat, so a solo developer running Claude Code against an API key pays only for the tokens a session actually burns, and a session that reads a large file and rewrites it across several turns can add up fast. Anthropic's own cost-management guidance is the best first-party anchor here: across enterprise deployments, the average developer spends around $13 on an active day and $150 to $250 a month, and 90% of users stay under $30 on their heaviest days.

Batch processing, for non-interactive jobs, cuts both input and output token prices in half on every model, but it does not apply to interactive Claude Code sessions, only to asynchronous API calls. For the per-token rates behind the Claude API itself, see best LLM APIs in 2026.

What ten developers pay next to Cursor, Copilot and Devin

Every figure below was checked on that vendor's own pricing page on September 25, 2026. At the individual entry tier, Claude Pro, Cursor Pro and Devin Pro all land on the same sticker, and GitHub Copilot Pro undercuts all three at $10/mo with 1,500 credits. At the heaviest individual tier, Claude's larger Max plan, Cursor's Ultra and Devin's Max all match that same sticker too, well above anything Copilot sells solo, since Copilot's top individual plan is Pro+ at $39/mo with 7,000 credits.

Seat math is where the four stop looking interchangeable. This table is only the invoice; the product choice, editor versus terminal agent, is covered in the Cursor review and Cursor vs GitHub Copilot.

Buyer	Claude Code	Cursor	GitHub Copilot	Devin
10 people, monthly seats	Team Standard, $250	Teams Standard, $400	Business, $190	Teams, $80 base + $400 in seats
10 people, cheaper interval	Team Standard, $2,400/yr	Teams Standard, $320/mo billed annually	No published annual rate	No published annual rate

Cursor's Teams Standard seat runs well above Claude's Team Standard seat, which is the trade-off between an editor built around Cursor's own model router and a terminal agent that shares Claude's subscription. Copilot's Business seat, at $19/user/mo, is the cheapest seat of the four, and plain code completions stay unlimited and off its credit meter on every paid tier, something neither Claude Code nor Cursor offers at any price. Devin's team plan adds an $80/mo platform fee on top of $40/mo per full seat, a line item Claude Code's Team plan does not charge. To see these plan prices against Cursor, including a 10-seat team cost, read Claude Code vs Cursor 2026.

Fees that never show up on the plan card

The subscription price is the seat, not the whole month. Usage credits let a Pro, Max, Team or Enterprise member keep working past their included allowance, billed at standard API rates once turned on, and a member with no spend limit set can run that meter with no ceiling. A team that never checks the spend report can be paying full seat prices and a growing usage-credit bill at the same time without anyone noticing until the invoice lands.

A long session that sits open for hours can burn far more of the weekly allowance than the actual work suggests, because Claude Code resends the whole conversation with every tool call, and a break longer than the prompt cache's lifetime forces a full, more expensive reprocess on the next message. Data residency, when an organization pins inference to the US, adds a 1.1x multiplier on every token category on the API, a fee that never shows up on a subscription plan card because it only applies to Console and cloud-platform billing.

Devshot data: this brief goes to 28,000 developers, and the audience file, updated September 2026, shows a 30% open rate and a 1.8% click rate. Plan and pricing changes for that list run in the Devshot brief.

Which plan is worth paying for

For one person, start on Pro and move to the smaller Max tier only after a real week has run out, not because a bigger number sounds safer. The larger Max tier is for people who already know their agent runs span most of the day; buying it up front on a hunch pays for headroom nobody is using yet.

For a team, Team Standard is the cheaper way in than API keys for most groups, and Premium seats belong only on the people who will actually burn 6.25x the allowance, not on the whole roster by default. If your team already knows its token spend from running the API directly, price that against Team seats before committing, since the API pricing docs let you model exact costs per model rather than estimating against a pooled seat allowance.

If you are still choosing the tool rather than the plan, how to use Claude Code covers the setup, and best AI coding assistants and AI for developers cover where it sits against the rest of the category. For the prompt-to-app end of the market instead of a terminal agent, best vibe coding tools is the right list, and if OpenAI's agent is the other option on your shortlist, see the Codex review. If you are weighing Claude Code against the free, open-source OpenCode agent, OpenCode vs Claude Code 2026 prices one month of agent work both ways.

How we checked these prices

Louis Corneloup, founder of Dupple and Toolradar, wrote this breakdown. We checked Anthropic's pricing page, the Claude Code product page, the API pricing docs, the Claude Code cost management guide, the Max plan article, the Team plan article and the Team and Enterprise Claude Code article on September 25, 2026.

Cursor figures come from cursor.com/pricing. Copilot figures come from GitHub's individual plans docs. Devin figures come from devin.ai/pricing. The 153 tool count is from Toolradar's live AI coding ranking that same month. More on how we publish is on the Devshot about page.

FAQ

How much does Claude Code cost in 2026?

There is no standalone Claude Code price. It rides on Pro, and above that on Max 5x or Max 20x. Team Standard seats are $25/user/mo, and Team Premium seats are $125/user/mo. Building against the API instead of a subscription bills per token with no monthly ceiling.

Is Claude Code free?

No. The Free Claude plan does not include Claude Code at all, only chat, web search and a handful of other features. The cheapest way to open a Claude Code terminal session is the Pro plan. That is different from Cursor and GitHub Copilot, which both offer a genuinely free tier of their own agent.

What is the difference between Max 5x and Max 20x?

Both include the exact same Claude Code features as Pro; the only difference is the size of the usage allowance. Max 5x, at $100/mo, gives five times Pro's per-session limit, and Max 20x, at $200/mo, gives twenty times that limit. Anthropic points the smaller tier at people who use Claude across varied tasks most days, and the larger one at daily users running long agent sessions.

How much does the Claude Code API cost?

Claude Sonnet 5, the model Anthropic points developers toward for most coding work, is $2/MTok for input and $10/MTok for output on the pricing docs. There is no monthly cap on the API. Anthropic's own cost-management guidance puts the average enterprise developer around $13 on an active day and $150 to $250 a month, with 90% of users staying under $30 on their heaviest days.

How much is Claude Code for a team of 10?

Ten Team Standard seats are $250/mo billed monthly, cheaper than ten Cursor Teams Standard seats at $400/mo and close to ten GitHub Copilot Business seats at $190/mo. Paying annually brings the Claude figure down to $2,400 for the year. Each seat carries its own allowance rather than a pooled team budget, so a mix of seat tiers fits a team where usage is uneven better than putting everyone on the same one.

Is Claude Code cheaper than Cursor?

At the entry tier, no, both individual plans land on the same monthly sticker. One step up, Claude Code's smaller Max tier costs more than Cursor's matching step, Pro+ at $60/mo, though both top out at the same price on their heaviest individual tier. On teams, Claude's Team Standard seat undercuts Cursor's Teams seat. Claude Code's price also buys chat and Claude Code from one pool, while Cursor's price is for the editor alone.

Cite this: Devshot, "Claude Code Pricing in 2026: Plans, Usage Limits and API Costs", September 2026.

Devshot is the daily brief on the tools and bills developers have to act on. Get it at devshot.co.

Some offers on this page may be paid placements or contain affiliate links.