Claude Code pricing: plans vs API, and which costs less

Is Claude Code free, and which plans include it?#

Claude Code itself costs nothing to install, but it needs an account that pays for the model. Anthropic’s setup docs say it requires a Pro, Max, Team, Enterprise or Console account, and that the free Claude plan doesn’t include it. You can also run it through Amazon Bedrock, Google Cloud’s Agent Platform (formerly Vertex AI) or Microsoft Foundry.

Your plan covers Claude Code in the terminal and in the VS Code and JetBrains extensions, and all of it draws on the same allowance as your chats. Max plans also come with monthly credits for the Claude API, but Anthropic’s help centre says those credits can’t be used in Claude Code or the Claude apps.

How do Claude Code usage limits work?#

Every plan has a usage limit that resets on a rolling five-hour session window, and paid plans add a weekly limit on top. Your activity in Claude on the web, desktop and mobile and in Claude Code all draws from the same pool. Anthropic doesn’t publish a fixed message or token count: how far a session goes depends on the length and complexity of the work, the model you choose and the features you use.

• Max gives 5x or 20x Pro’s usage per five-hour session. Its weekly limit resets at a fixed time assigned to your account, shown in Settings > Usage.
• Fable models run on pay-as-you-go usage credits on Pro and on Team Standard seats. On Max and Premium seats you can spend up to 50% of your weekly limit on them.
• Opus or Sonnet limits apply only to that model family, so /model can switch you to another family. Session and weekly limits cover every model.

Run /usage to see your plan’s usage bars and when each window resets. When you hit a limit, you can wait for the reset (Claude Code can continue the interrupted task for you afterwards), upgrade, or turn on usage credits with /usage-credits. Usage credits are billed at standard API rates, and you can set a monthly spend limit. The usage limit reached page covers each message and fix.

How much does Claude Code cost on the API?#

With a Console account or API key, you pay for every token at list price, with no plan limits. The first time you sign in with a Console account, Anthropic creates a “Claude Code” workspace for tracking, and you can set workspace spend limits. /usage shows an estimated cost for the session; the Console’s usage page is the authoritative bill.

The cost is easiest to judge per step. Claude Code works in a loop: each time Claude reads a file, runs a command or replies, it sends a new request carrying the whole conversation so far. Prompt caching bills the repeated part at a much cheaper rate, so a typical step is a large cache read, a small cache write and some output. Every figure below uses this moderate step:

• 100,000 tokens of context read from the cache (a session that’s cleared between tasks)
• 2,000 new tokens written to the cache (tool results and your message)
• 1,200 output tokens, including thinking, which is billed as output

Opus 5.5 is Claude Code’s default model on Pro, Max, Team, Enterprise and the API, and its output costs twice as much as Sonnet 5.5’s. Switching to Sonnet for routine work, with /model sonnet, is the quickest way to cut an API bill.

Is a Claude plan cheaper than the API for Claude Code?#

Yes, unless you use it rarely. A plan is a flat price, while the API grows with every step. With our moderate step on Opus 5.5, the API passes Pro’s price at about 17 steps a working day. At the pace we measured (about 112 steps per active hour), that’s roughly 9 minutes of agent work a day.

That last condition matters. Anthropic describes plan limits in relative terms (“5x or 20x more usage than Pro”), not tokens, so nobody outside Anthropic can say exactly how many steps a plan allows. What the chart shows is the price of the same work at API rates. If you regularly hit Pro’s limit, the useful comparison is Max against the API, not Pro against the API.

What would light, daily and heavy use cost?#

Here are three usage levels, priced with the moderate step above. A step is one request, and a single task can take many of them. Daily use works out to about 2.7 hours of active agent work a day at our measured pace.

• Light use costs $17.28 a month on Opus 5.5, less than Pro. On Sonnet 5.5 it’s $8.64. If you code with Claude a couple of times a week and don’t use Claude chat, the API is a fair choice.
• Daily use costs $356.40 on Opus 5.5, 3.6 times the price of Max 5x, and $178.20 on Sonnet 5.5. If a plan’s limits cover this much work, it costs a fraction of the API.
• Heavy use at a full working day’s pace costs $1,117 a month on Opus 5.5. This is the kind of all-day use Max 20x is aimed at.

Real sessions are often bigger than our moderate step. The session we measured on 2026-10-08 was never cleared, so its median request carried 426,571 tokens of context. Its 940 requests would cost about $121.52 for that single day at Opus 5.5 list prices, and $1,792 without prompt caching. Keeping context small is the subject of our guide to reducing Claude Code token usage.

How much do developers actually spend on Claude Code?#

Anthropic publishes one figure, for organisations on API-style billing. Its cost documentation says: “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.” It adds that costs vary widely with the model, codebase size and habits such as running several instances, and suggests a small pilot to set a baseline.

Our daily example works out to $16.20 per working day on Opus 5.5, within the $30 a day that Anthropic says 90% of users stay below. To price your own token mix with caching and batch discounts, use the LLM cost calculator; for the method behind estimates like these, see how to estimate LLM API costs.

Claude Code for teams: Team, Enterprise or the Console?#

Anthropic’s docs recommend Claude for Teams or Enterprise for most organisations, with the Console for teams that prefer API billing. What differs is how usage is capped and reported:

• Team: Claude Code is included with every seat. Each member’s allowance resets on a five-hour and a weekly window and is shared with Claude chat. Standard seats give more usage than Pro; Premium seats give 5x Standard.
• Enterprise: a seat fee plus usage billed at API rates, billed annually. On usage-based plans there are no per-seat limits; admins set spend limits for the organisation, groups or members.
• Console: per-token billing to your API organisation, workspace spend limits, and per-user spend and accepted lines in the Console dashboard or the Claude Code Analytics API.

On Team and seat-based Enterprise plans, admins can also turn on usage credits so members keep working past their allowance. Anthropic’s consumption guide tells admins to budget more for a coding seat than a chat seat, because each Claude Code turn carries file contents, tool calls and multi-step reasoning.

Using Claude Code through Amazon Bedrock or Google Cloud#

Claude Code can also run on your company’s AWS or Google Cloud account. You pay per token at your cloud provider’s rates, and spending shows in your cloud billing console. Anthropic’s analytics dashboards don’t cover this usage, so per-user costs come from OpenTelemetry or a gateway. Run claude, choose 3rd-party platform at the login prompt, and a wizard sets it up. Or set the variables yourself:

# Amazon Bedrock
export CLAUDE_CODE_USE_BEDROCK=1
export AWS_REGION=us-east-1

# Google Cloud's Agent Platform (formerly Vertex AI)
export CLAUDE_CODE_USE_VERTEX=1
export CLOUD_ML_REGION=global
export ANTHROPIC_VERTEX_PROJECT_ID=YOUR-PROJECT-ID

• Pin your models. On both clouds the default primary model is Opus 5.5, so an unpinned setup is billed at Opus rates. Set ANTHROPIC_MODEL, or the ANTHROPIC_DEFAULT_OPUS_MODEL and ANTHROPIC_DEFAULT_SONNET_MODEL variables, to the model IDs your account has enabled.
• Expect shorter caching. The prompt cache lives with your cloud provider and lasts five minutes by default. ENABLE_PROMPT_CACHING_1H=1 requests an hour, at a higher cache-write price.
• Some features differ. For example, the web search tool isn’t available on Bedrock.