Published December 04, 2025 · Updated August 27, 2026
What are Claude Code’s token limits?
Earlier in 2026, Anthropic began describing Claude Code’s token limits in more relative terms rather than as fixed token counts. Claude has two types of limits: length limits and usage limits. The key distinction is that length limits determine how long a single conversation can become, whereas usage limits determine how much you can use Claude overall across your conversations. In other words, length limits measure the size and complexity of one conversation, while usage limits measure total activity over time.
Claude Code length limits
Claude’s context window size is 200K tokens across all models and paid plans, except for Enterprise plans, which have a 500K context window on some models. Once a conversation or codebase exceeds that window, Claude may lose access to earlier details, which can make long debugging sessions, large refactors, and multi-file projects harder to manage.
Claude Code usage limits
Claude Code operates on a 5-hour rolling window that begins with your first message in a session. Rather than publishing a fixed token allocation for each plan, Anthropic now describes usage capacity relative to the Pro plan, with Max tiers providing higher usage per session.
| Claude Plan |
Monthly Cost |
Usage Capacity per 5-Hour Session |
Relative to Pro |
| Pro |
$20 |
standard |
1x baseline |
| Max 5x |
$100 |
5x Pro capacity |
5x |
| Max 20x |
$200 |
20x Pro capacity |
20x |
Claude plan monthly costs and relative usage capacity per 5-hour session.
Note: Usage on Pro and Max plans is shared across claude.ai, Claude Code, and Claude Desktop. Messages or activity in any one of those surfaces count against the same usage pool, which is why Claude Code users may hit usage limits sooner than expected if they are also using Claude elsewhere. Actual usage varies based on factors such as conversation length and complexity, the model you choose, and the features you use, so these plan multipliers shouldn’t be interpreted as fixed token or prompt allowances.
Enterprise clients have a different usage model. On Anthropic’s current usage-based Enterprise plan, the seat fee covers access to Claude, Claude Code, and Cowork, while usage is billed separately based on actual token consumption at standard API rates. Unlike Pro, Max, Team, and legacy seat-based Enterprise plans, usage-based Enterprise has no included token allowance or per-seat usage limits. Admins can instead control consumption by setting spend limits at the organization and individual user levels.
Since August 2025, weekly limits sit on top of these 5-hour windows. The current structure is one weekly cap that applies across all models, plus a separate weekly cap that applies specifically to Sonnet usage. This was a response to a small number of users who were, as Anthropic put it, consuming resources at unsustainable rates.
In March 2026, Anthropic temporarily adjusted Claude Code’s 5-hour session limits during peak hours, causing Free, Pro, and Max users to move through their session limits faster on weekdays between 5am–11am PT. However, Anthropic reversed this change for Pro and Max accounts on May 6, removing the peak-hours limit reduction. At the same time, Anthropic doubled Claude Code’s 5-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans.
What happens when you hit your Claude Code usage limit?
Hitting a Claude Code usage limit doesn’t necessarily mean you have to stop working. What happens next depends on your plan and how Claude Code is configured.
For Pro and Max users, you can wait for your usage limit to reset, upgrade to a higher-usage plan, or enable usage credits to continue working beyond your plan’s included allowance. Once enabled, additional usage is charged at consumption-based rates. Users can also switch to pay-as-you-go usage through a Claude Console account for more intensive coding workloads.
For Team and seat-based Enterprise plans, organizations can enable usage credits so developers can continue working after reaching their included limits. Usage-based Enterprise plans work differently: there are no per-seat usage limits, and consumption is billed at API rates.
For engineering leaders, this means hitting a Claude Code limit is increasingly a cost-management issue rather than simply an access issue. Teams may be able to keep coding past their included allowances, but doing so can introduce variable spend that needs to be tracked and governed.
How different models affect Claude Code token limits
Claude Code usage depends on several factors, including the length and complexity of your conversations, the features you use, and your selected model and effort settings. Model choice directly affects how quickly Claude Code usage is consumed. Claude Code model pricing is based on input and output tokens, as summarized in the following table:
| Claude Code Model |
Current Model Tier |
Input Token Price |
Output Token Price |
Total Cost for 1M Input + 1M Output |
Relative Cost Across Model Tiers |
Best For |
| Claude Opus |
Opus 5 |
$5 / 1M tokens |
$25 / 1M tokens |
$30 |
5x Haiku |
Complex reasoning, large codebase work, high-autonomy agentic coding |
| Claude Sonnet |
Sonnet 5 |
$2 / 1M tokens |
$10 / 1M tokens |
$12 |
2x Haiku |
Everyday Claude Code use, refactoring, debugging, balanced speed and quality |
| Claude Haiku |
Haiku 4.5 |
$1 / 1M tokens |
$5 / 1M tokens |
$6 |
1x baseline |
Lower-cost tasks, fast iterations, simpler coding assistance |
Claude Code model tiers, token pricing, relative costs, and recommended use cases.
Across all three models, output tokens are the bigger cost driver, with each model’s output tokens costing 5x more than its input tokens. And, for the same number of input and output tokens, Sonnet costs 2x more than Haiku, while Opus costs 5x more than Haiku. Practically speaking, that means heavy use of Opus will exhaust your Pro/Max allocation much faster than Sonnet or Haiku usage. If you’re running complex, multi-file agentic workflows with Opus, you'll hit your limits much sooner than you might expect.
A note on comparing token usage across models: Sonnet 5 introduced an updated tokenizer, which means the same input can translate into more tokens than it did with Sonnet 4.6. Anthropic estimates that the same input can map to roughly 1.0–1.35x as many tokens, depending on the content type. That means raw token counts aren't necessarily an apples-to-apples comparison across model generations, even when the underlying workload stays the same.
How effort levels affect Claude Code token usage
Model choice isn’t the only factor that determines how quickly you consume Claude Code usage. Reasoning effort also matters. Claude Code lets users adjust how much computational effort Claude applies to a task, trading off deeper reasoning against latency and token consumption.
Higher effort levels can improve performance on complex coding and agentic tasks, but they also consume more tokens and can cause users to hit usage limits faster. Lower effort levels can be more efficient for simpler tasks where extended reasoning isn’t necessary. Anthropic describes this as a tradeoff between more thinking and lower latency and fewer usage-limit hits.
For engineering teams, that means understanding Claude Code consumption increasingly requires looking at both the model and the effort level being used. Two developers using the same model for similar workloads can consume meaningfully different amounts of their usage allowance depending on how much reasoning effort they apply.
How advanced features affect Claude Code token limits
Claude has numerous types of advanced features that can greatly increase token usage. There are two worth noting:
Agent Teams: In February 2026, Anthropic released Agent Teams. This multi-agent capability is now a built-in part of Claude Code, and it can significantly increase the number of tokens software engineers use during a session. Agent teams run multiple Claude Code instances at once, with each instance maintaining its own context window. As a result, token consumption grows based on how many teammates are active and how long they continue running. Anthropic notes that agent teams can consume about 7x more tokens than standard sessions when teammates operate in plan mode.
Dynamic Workflows: In May 2026, Anthropic released dynamic workflows (for those on Claude Enterprise plans), and they became available and turned on by default on June 8, 2026. Dynamic workflows can further expand token consumption by turning a single request into a scripted, multi-agent execution. Instead of Claude handling the task turn by turn in one conversation, a workflow can fan work out across dozens or even hundreds of subagents, each performing its own model calls and tool use. Anthropic notes that workflow runs can use meaningfully more tokens than completing the same task through a standard conversation, and those runs count against the organization’s usage and rate limits.
How to reduce Claude Code token usage
Because Claude Code’s token usage scales with the amount of context it processes, keeping that context focused can help developers get more from their usage limits. Anthropic recommends several ways to reduce unnecessary token consumption:
- Clear context between unrelated tasks. Use
/clear when moving to a new task so Claude doesn’t continue processing irrelevant conversation history with every subsequent message. - Compact long-running conversations. Use
/compact to summarize the conversation while preserving the information needed to continue working. Claude Code also automatically compacts conversations as they approach the context limit. - Use the right model for the task. Sonnet is suitable for most coding tasks, while Opus can be reserved for work that requires more complex reasoning. Simpler tasks can also be delegated to Haiku-powered subagents.
- Watch what’s consuming context. The
/context command shows what is taking up space in the current context window, making it easier to identify oversized instructions, tools, or other sources of unnecessary context. - Limit unnecessary tool context. Unused MCP servers can add to context consumption. Anthropic recommends disabling servers you aren’t actively using and using CLI tools where appropriate.
These practices can help developers stretch their Claude Code usage further, but optimizing for fewer tokens shouldn’t be the goal in isolation. For engineering organizations, the more useful question is whether the tokens being consumed are producing valuable outcomes, such as completed work, merged PRs, and faster delivery.
Claude Code token limits: What engineering leaders should know about AI coding costs
AI coding tools like Claude Code are more widely used in software development than ever—and costs have climbed just as fast. Yet, that spend remains hard to manage: consumption-based pricing is unpredictable, actual limits are opaque, and the link between AI usage and engineering outcomes is murky.
Anthropic also provides native analytics for tracking Claude Code usage, contribution, and cost, including sessions, token consumption by model, commits, pull requests, and estimated cost per user. These metrics are useful for understanding adoption and spend, but they don't show what happens to AI-assisted work after it leaves the tool. For a deeper look at the available data, ingestion options, and where tool-level telemetry stops, read our article on Claude Code analytics.
To see what your organization's AI spend is actually producing, start with The Field Guide to Measuring Token Efficiency in AI Engineering, which lays out the metrics worth tracking so you can make decisions grounded in your own data. From there, see how Token Intelligence traces AI token consumption to what it delivers across your people, teams, and outcomes—so you know what's productive, what's wasteful, and what to fix.
Frequently asked questions about Claude Code token limits
What is the Claude Code context window size?
Claude Code's context window is 200K tokens across all models and paid plans. Enterprise plans get a 500K window on some models. Once a conversation or codebase exceeds the window, Claude can lose access to earlier details, making long debugging sessions and large refactors harder.
How many tokens do you get with Claude Pro vs. Max?
Anthropic no longer publishes fixed token allocations for Pro and Max plans. Instead, it describes usage capacity relative to Pro: Pro ($20/month) is the baseline, Max 5x ($100/month) provides 5x Pro usage per 5-hour session, and Max 20x ($200/month) provides 20x Pro usage. Actual usage varies based on factors such as conversation length and complexity, model choice, and the features you use. Usage is also shared across claude.ai, Claude Code, and Claude Desktop.
Does Claude Code have weekly limits?
Yes. Since August 2025, weekly caps sit on top of the 5-hour windows: one weekly cap across all models, plus a separate weekly cap specific to Sonnet usage. Anthropic added them in response to a small number of users consuming resources at unsustainable rates.
How much does Claude Code cost per developer?
About $6 per developer per day on average, with 90% of users below $12/day. Team deployments on the API with Sonnet typically run roughly $100–$200 per developer per month, depending on usage intensity.
Why does Opus burn through Claude Code limits faster than Sonnet?
For the same volume of input and output tokens, Opus costs 5x Haiku while Sonnet costs 2x Haiku. Across all three models, output tokens cost 5x more than input tokens. Heavy Opus use on complex, multi-file agentic workflows can therefore exhaust your Pro or Max usage faster than Sonnet or Haiku. Reasoning effort also matters: higher effort levels consume more tokens and can cause developers to hit usage limits faster, even when using the same model.
What is Claude Code's pricing per million tokens?
Claude Opus 5 costs $5 per 1M input tokens and $25 per 1M output tokens. Claude Sonnet 5 costs $2 per 1M input tokens and $10 per 1M output tokens. Claude Haiku 4.5 costs $1 per 1M input tokens and $5 per 1M output tokens. Opus is best suited to complex reasoning and large-codebase work, Sonnet to everyday coding, refactoring, and debugging, and Haiku to faster, lower-cost tasks.
Do Claude Code Agent Teams use more tokens?
Yes, significantly. Agent Teams (released February 2026) run multiple Claude Code instances at once, each with its own context window, so consumption scales with how many teammates are active. Anthropic notes Agent Teams can use about 7x more tokens than standard sessions when teammates run in plan mode.
What are Claude Code dynamic workflows and how do they affect token usage?
Dynamic workflows (Enterprise plans, on by default since June 8, 2026) turn a single request into a scripted, multi-agent execution that can fan work across dozens or hundreds of subagents, each making its own model calls. They use meaningfully more tokens than the same task in a standard conversation, and runs count against the org's usage and rate limits.
Did Anthropic remove Claude Code’s peak-hour limits?
Yes, for Pro and Max users. In March 2026, Anthropic temporarily reduced effective Claude Code session limits during peak weekday hours. On May 6, Anthropic removed that peak-hours reduction for Pro and Max accounts. It also doubled Claude Code’s 5-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans. Weekly and other usage limits can still apply.
What metrics should you track to manage Claude Code spend?
Track token consumption and cost by model, user, and team to understand where Claude Code spend is going and identify unusual usage patterns. Anthropic's native analytics can also provide visibility into adoption and contribution metrics such as sessions, commits, and pull requests. But cost and usage should ultimately be connected to engineering outcomes, including cycle time, deployment frequency, quality, rework, and incidents, to understand whether increased AI spend is actually producing value.
What happens when you hit your Claude Code usage limit?
What happens depends on your plan. Pro and Max users can wait for their usage limit to reset, upgrade to a higher-usage plan, enable usage credits to continue at consumption-based rates, or switch to pay-as-you-go usage through a Claude Console account. Team and seat-based Enterprise organizations can also enable usage credits, while usage-based Enterprise plans have no per-seat usage limits and instead bill consumption at API rates.
.webp)
.webp)
.webp)