How much does Claude Code cost in 2026?

Claude Code is included in every paid Claude plan. Individual plans run from $20 a month for Pro to $200 a month for Max 20x. Team seats start at $20 per seat per month billed annually, and Enterprise is publicly advertised at $20 per seat per month plus usage at API rates. Developers who’d rather skip a subscription can pay per token through the Claude Console. Which option costs least comes down to how heavily each developer uses Claude Code and how many developers need access.

But the sticker price is only part of the answer.

Some older Claude models cost more per token than their newer replacements, a cheaper model can end up costing more per task, and two developers on the same plan can spend very different amounts. This guide covers plan and API pricing, per-developer cost benchmarks, how to choose a plan, and how to judge whether Claude Code is worth what you pay. For how the 5-hour and weekly usage windows work, see our guide to Claude Code token limits.

How much does Claude Code cost on each plan?

Claude Code is available on Pro, Max, Team, and Enterprise. Anthropic’s free plan does not include Claude Code.

Note: Max prices are for web subscriptions. If you subscribe through the mobile app, you may pay a different price.

Max plans come with monthly API credits. You can claim them into a linked Claude Console organization once you’ve been on the plan for seven days. They expire at the end of each billing cycle, and you can’t use them for interactive Claude Code sessions or to cover subscription overages.

Team plans get pooled API credits, too: $20 per Standard seat and $100 per Premium seat, up to $500 a month. They cover claude -p and Agent SDK runs that use an API key from the linked organization. Runs signed in with a Claude subscription draw from your plan’s usage instead.

How do you know if Claude Code is worth the cost?

You can determine if Claude Code is worth the cost if you can tie its spend to work that ships: merged pull requests, resolved tickets, and released features. We call that measure cost per outcome, and it is the number an engineering leader needs before renewing, expanding, or capping seats.

The invoice alone can’t make that call. A large AI bill can mean a team is stuck in expensive loops or shipping faster than ever, and it often looks the same either way. Faros CEO Vitaly Gordon described one Faros engineer spending $20,000 on tokens in a month. In his account, nobody could say whether that called for a promotion or a stern conversation, because the cost number carried no information about what value it produced.

Spend also diverges between developers on the same plan, for reasons the bill doesn’t show: model choice, the context the agent gets, and how each developer splits work with it. Task difficulty matters too. A high cost per outcome can reflect a hard repository or work that never needed AI, and reviewing the underlying pull requests, task complexity, and delivery outcomes helps explain which. That is why we urge engineering leaders to move from token maxxing to outcome maxxing, starting with visibility into what each seat’s spend actually produced.

Does Claude Code share usage limits with Claude chat?

Yes. On Pro, Max, and Team plans, Claude Code draws from the same usage pool as Claude on web, desktop, and mobile. That pool resets on a rolling 5-hour window, and paid plans add weekly limits. Anthropic publishes no universal fixed message count, because usage depends on conversation length, model, and features used. Model-specific limits can also apply, including a separate weekly Fable limit on Max.

What happens when you run out of included usage in Claude Code?

On plans with included usage, eligible users can turn on usage credits, which bill further Claude Code and chat usage at standard API rates, subject to organizational controls where applicable. On Pro and Max, you prepay a balance and can set a monthly spend cap. Usage credits are separate from the monthly API credits included with Max, and those API credits can’t be applied to usage credits in Claude Code.

How does pay-as-you-go API pricing work for Claude Code?

Without a subscription, Claude Code bills by API token consumption through the Claude Console. Users pay per million tokens (MTok) of input and output, at a rate set by the model. Cache writes and cache reads have separate rates.

What do the newest Claude Code models cost?

Opus 5.5, Claude Code’s default model, costs $4 per million input tokens and $20 per million output tokens, with Sonnet 5.5 and Haiku 5.5 costing less and Fable 5.1 costing more.

What other models are available in Claude Code, and how much do they cost?

Claude Code also supports older models. You can switch to one by entering its model ID with /model, as long as your provider and account allow it. Some cloud providers point Claude Code’s shortcut names, like “opus” and “sonnet,” to these older versions by default, so you may be on one without realizing it.

These are list prices from Anthropic’s API pricing, but a few things change what you actually pay:

• Caching: Claude Code caches prompts automatically. Reused context, like system prompts, is billed at the cache-read rate. Writing or rebuilding the cache costs 1.25x the input rate for a five-minute cache, or 2x for a one-hour cache.
• Thinking tokens are billed as output tokens.
• Fast mode for Opus 5.5 costs twice the standard price.
• US-only inference adds 10% to all token charges on supported models and platforms.
• Cloud providers: On Amazon Bedrock, Google Cloud, and Microsoft Foundry, you’re billed through your cloud account at that provider’s prices, including its regional pricing.

Developers beware: some older models cost more than their newer replacements. Opus 5 and Opus 4.8 cost $5 per million input tokens, while Opus 5.5 costs $4. If you or your team are pinned to an older model, you may be paying more per token. Keep in mind that tokenization differs between model generations, so the same text can come out to a different number of tokens. A lower per-token price doesn’t guarantee a lower bill.

Will using a cheaper model lower your Claude Code costs?

Not necessarily. Per-token price is one input to cost per task, not the whole of it. A cheaper model that needs more attempts, longer sessions, or more human review can cost more overall than a pricier one that gets the work right the first time. We call this pattern the hidden model tax.

The reverse also holds: the most capable model isn’t the right default for every task. Anthropic’s guidance is that Sonnet handles most coding tasks well and costs less than Opus, with Opus reserved for complex architectural or multi-step work.

So how do you know which model is worth the cost?

The right answer depends on your tasks and your codebase, and testing various model and harness combinations on your own engineering work provides stronger evidence for your environment. When we did this experiment at Faros, we ran 211 of our own engineering tasks through seven AI coding routes. Claude Code running the open model GLM-5.2 received a higher average model-judge score than the tested Claude Code + Opus 4.8 and Codex + GPT-5.5 routes, at roughly half their projected token cost per task. 

Running that kind of test by hand doesn’t scale, especially since new models ship faster than teams can re-evaluate them. Faros’s Time Machine automates this evaluation process. It builds benchmarks from your own merged pull requests, replays those tasks on alternative models and harnesses in isolated workspaces, and scores each route on cost and quality against what actually shipped. Teams can enforce the resulting routing policies through Faros’s gateway or their own, and re-evaluate routes as tools and code change.

Which Claude Code plan should you choose?

The right plan depends on how heavily Claude Code gets used and whether it’s for one developer or a team. Hours at the keyboard aren’t a reliable measure of usage on their own. Model choice, context size, parallel agents, and task complexity all affect token consumption and cost. Anthropic’s per-developer averages, covered below, help estimate spend, but they won’t show which plan covers a given workload.

Should an individual developer choose Claude Code Pro, Max, or API billing?

Pick by how often you use Claude Code: Pro for occasional use, Max if you use it every day, and API billing if your usage is unpredictable or automated.

• Pro fits occasional use, such as learning Claude Code or running it a few times a week.
• Max 5x or Max 20x fits frequent users who need higher allowances. Anthropic positions Max for longer, uninterrupted sessions with higher output limits and priority access at high-traffic times, but it does not guarantee uninterrupted all-day coding.
• API billing fits variable or automated workloads, such as CI jobs, where per-token billing tracks actual use.

The flat Max price is useful when compared against Anthropic’s reported average of $150 to $250 per developer per month, but that usage-cost benchmark does not establish that Max covers the same workload. A heavy user on a $100 or $200 plan pays a fixed subscription fee, but still works within the plan’s session and weekly limits unless usage credits are turned on.

How should an engineering team choose between Standard seats, Premium seats, and Enterprise?

Choose by how heavily each developer uses Claude Code and which security and admin controls your organization requires.

• Standard seats ($20 per seat per month billed annually, $25 monthly) fit developers who use Claude Code occasionally alongside chat. They include more usage than Pro.
• Premium seats ($100 per seat per month billed annually, $125 monthly) fit developers who run Claude Code daily, with 5x the usage of a Standard seat. The Team plan lets you mix both seat types across 2 to 150 people.
• Enterprise (publicly advertised at $20 per seat per month billed annually, plus usage at API rates, 20-seat minimum) fits organizations that need SCIM, audit logs, role-based access, custom data retention, or a HIPAA-ready offering. 

In practice, Team suits organizations of 150 people or fewer whose developers work within seat allowances. Enterprise suits larger organizations, and any that need SCIM, audit, and compliance controls.

How much does Claude Code cost per developer per month?

Across enterprise deployments, Anthropic reports that Claude Code costs about $150 to $250 per developer per month and around $13 per developer per active day. For 90% of users, costs stay below $30 per active day. To estimate your own team’s spend, Anthropic recommends starting with a small pilot group to set a baseline.

Those averages are a useful starting point for a budget, but they hide how unevenly AI spend is spread. Some developers incur much higher costs, so a uniform per-developer cap may constrain heavier users even when overall spending stays within budget. Faros recommends tracking AI coding costs and shipped outcomes more granularly, so engineering leaders can achieve a higher degree of AI coding ROI.

Track your Claude Code costs with Faros

Claude Code is one part of your AI coding bill. Faros brings costs across your AI coding tools into one view and connects that spend to engineering outcomes. Track costs by developer, team, and task, see what the investment produces, and use that evidence to guide model choices and spending decisions.

Connect with our team to learn more.