You paste last month's Google Ads export into Claude, then the Meta
one, ask for a CPA comparison, and spend a few turns on headline ideas
for the winning campaign. Before lunch, Claude tells you you've reached
your limit, and you're not sure what you spent it on.
Long threads, pasted CSVs, and tool results all contribute to Claude
usage. This guide explains the plan limits, a measured synthetic
reporting session, and ways to reduce repeated work before moving
recurring ad operations outside the chat window.
How Claude usage limits work
**Claude usage limits [cap how much work you can do with Claude over
time](https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work),
not a fixed number of messages.** Free and paid subscription plans get a
session limit that resets every five hours; paid ones add weekly limits,
and claude.ai, Claude Desktop and Claude Code share one pool.
Usage-based Enterprise is the exception: it bills by consumption.
Session limits, weekly limits, and shared usage affect the same workflow
in different ways:
- The 5-hour session limit. Free, Pro, Max, and Team plans all have
one. When you burn through it, you wait for the reset, switch on usage
credits (paid plans only), or upgrade.
- Weekly limits. Paid subscription plans add a weekly usage limit on
top, which resets at a fixed time assigned to your account. A heavy
Monday can leave you short on Friday even if every session resets in
between.
- One pool across Claude surfaces. A CSV you analyze in Claude
Desktop in the morning and a Claude Code session in the afternoon come
out of the same allowance.
- Weight, not count. What a message costs depends on the length and
complexity of the conversation, the features and tools you use, the
model, and the effort level.
Anthropic reserves the option to [impose additional weekly, monthly,
model or feature caps on
Pro](https://support.claude.com/en/articles/8325606-what-is-the-pro-plan).
Limits can also change over time. On May 6, 2026, for example,
[Anthropic doubled Claude Code's five-hour rate
limits](https://www.anthropic.com/news/higher-limits-spacex)
for Pro, Max, Team and seat-based Enterprise plans and removed the peak
hours limit reduction on Claude Code for Pro and Max. Treat any number
you read as a snapshot and check Settings \> Usage for your own.
Usage limits by plan: Free, Pro, Max, Team and Enterprise
Yes, Claude Pro has usage limits, as do the other consumer and Team
plans. There is [no fixed message
allowance](https://claude.com/pricing): session use depends
on conversation length, complexity, model, and features. The tiers below
use per-session multiples rather than guaranteed message counts.
Last checked against Anthropic's Help Center and pricing page on
October 12, 2026.
Plan Usage per 5-hour Weekly limit Usage credits
session
- ----------------- ----------------- ----------------- -----------------
Free Baseline; varies None listed No
with demand (Anthropic says
paid plans add
weekly limits)
Pro (\$20/month, At least 5x Free Yes, across all Yes
or \$17/month models
billed annually)
Max 5x 5x Pro Yes, plus a Yes
(\$100/month) separate Fable
weekly limit
Max 20x 20x Pro Yes, plus a Yes
(\$200/month) separate Fable
weekly limit
Team, Standard 1.25x Pro Yes, per member Yes, enabled by
seat an Owner
Team, Premium 6.25x Pro (5x Yes, per member Yes, enabled by
seat Standard) an Owner
Enterprise No plan or No Not applicable:
(current seat-level limit all usage billed
usage-based plan) at API rates
Read the allowances with these plan-specific conditions:
- Free. The [Free plan's
limit](https://support.claude.com/en/articles/8114491-get-started-with-claude)
resets every five hours, and the number of messages "will vary based
on demand".
- Pro. Under Claude Pro usage limits, the number of messages varies
with message length, attached files, the length of your current
conversation, and the model or feature you use. Claude Fable models
run on usage credits rather than the Pro allowance.
- Max. [Claude Max plan usage
limits](https://support.claude.com/en/articles/11049741-what-is-the-max-plan)
are Pro's per-session allowance times 5 or 20. On Max, up to 50% of
the weekly limits can go to [Fable
models](https://support.claude.com/en/articles/15424964-claude-fable-models-on-your-plan).
- Team. [Team plan
limits](https://support.claude.com/en/articles/9266767-what-is-the-team-plan)
are per member, so one heavy user doesn't drain a colleague's
allowance.
- Enterprise. The current [Enterprise
plan](https://support.claude.com/en/articles/9797531-what-is-the-enterprise-plan)
has no plan- or seat-level usage limits: a \$20 monthly seat, billed
annually, plus all usage at API rates, with admin spend limits. Older
seat-based Enterprise plans keep their limits.
Claude Code is included in every paid plan and draws on the same limits
as chat, so for Claude Code usage limits, read the same rows.
Usage limits vs length limits (context window)
Claude has two types of limits that are easy to confuse. **A usage limit
caps how much you use Claude across all your conversations; a length
limit caps how long one conversation can get**. Length limits come from
the context window, the amount of information Claude can work with in a
single chat.
Usage limit Length limit
- ----------------------- ----------------------- -----------------------
What it caps Total use over five One conversation
hours and a week
What runs it down Long threads, files, Everything in the
tools, model, effort current chat
When you hit it Wait for the reset, use Start a new
usage credits, or conversation or use a
upgrade project
On paid plans, Anthropic says the newest models support up to a 1M token
context window and others 500K or 200K; its [per-model context window
table](https://support.claude.com/en/articles/8606394-how-large-is-the-context-window-on-paid-claude-plans)
lists 1M for Opus 5.5, Sonnet 5.5 and Haiku 5.5 in chat. The Team plan
overview still mentions a 200K context window, so Team users should
check the per-model table for the model they use.
With code execution enabled, Claude can summarize earlier messages as
the chat approaches its context limit. Conversations that trigger this
automatic context management consume more usage. A larger context window
lets the thread continue; it does not make the additional history free
to process.
Why marketers hit Claude's limits so fast when running ads
Consider this ad-ops workflow: summarize an export, add a second export,
compare platforms and iterate on copy in the same thread. We ran five
tasks with raw exports and compared them with campaign-level summaries
plus a separate copy conversation, recording tokens read and written per
turn.
Example session (measured): one long thread with raw exports pasted.
Turn Request Tokens Claude Tokens Claude
read wrote
- ----------------- ----------------- ----------------- -----------------
1 Paste a 30-day 29,202 12,377
Google Ads
export, summarize
by campaign
2 Paste the Meta 52,766 8,346
export, compare
CPA across both
platforms
3 Three headline 61,171 522
ideas for the
best Google
campaign
4 Make them shorter 61,731 746
and add a number
5 Re-run the 62,546 6,687
comparison for
the last 7 days
Total 267,416 28,678
The same five tasks, done with campaign-level summaries pasted instead
of raw rows and the headline work moved to a new conversation, read
64,238 tokens and wrote 2,850. That's about a quarter of the tokens
read and a tenth of the tokens written. The headline request on its own
read 15,197 tokens in a fresh chat against 61,171 at turn 3 of the long
thread.
*Method: Claude Code 2.1.296 (claude -p), Sonnet 5.5, default effort,
tools off, one run on 11 October 2026, with synthetic data shaped like
daily campaign exports (Google: 360 rows; Meta: 300). "Read" counts
input and cached tokens for the main model, including Claude Code's own
instructions. Anthropic doesn't publish how tokens or cache hits
convert into plan usage, so treat this as relative workload, not a meter
reading.*
Raw exports and long reporting threads
The raw workflow repeatedly sent the exports and asked Claude to
aggregate daily rows. Claude Code resends conversation history on each
turn, so later questions in our run carried about 45,000 CSV characters
even when the question only concerned a headline.
The output side surprised us more. Of the 12,377 tokens Claude wrote on
turn 1, 11,849 were thinking, spent adding up 360 daily rows before it
could show a 12-row table (its answer says it totaled them by hand). A
campaign-level summary, the kind [a unified cross-platform
report](https://synterai.com/blog/unified-ads-reporting-dashboard)
gives you, does that arithmetic before Claude ever sees the data.
Connectors and MCP tools that return too much data
MCP [tool output becomes input to
Claude](https://platform.claude.com/docs/en/agents-and-tools/tool-use/overview)
through tool_result blocks. Connecting an ad account changes how the
data arrives, not whether the model processes it. Claude Code [warns
above 10,000 tokens](https://code.claude.com/docs/en/mcp)
for one MCP result and uses a default maximum of 25,000 tokens, saving
an oversized result to a file.
So ask what any ads connector sends back. A server that returns every
row the platform API offers hands Claude the same job as a pasted CSV.
Synter, our own product, is one example: our [Google Ads and Meta
reporting tools](https://docs.synterai.com/mcp/tools)
return impressions, clicks, spend, and conversions per campaign for the
date window you choose, so a 30-day question doesn't have to arrive as
360 daily rows. That output still counts toward your Claude usage like
any other tool result, and we didn't measure our tools here, so we
won't put a number on the difference. For the architecture behind it,
see [how generic ad MCPs are
built](https://synterai.com/blog/why-we-dont-use-ad-platform-mcps)
and the comparison of [a single-platform MCP vs an operating
layer](https://synterai.com/blog/claude-meta-mcp-vs-synter).
Copy iteration in one endless chat
Headline rounds are light on their own and heavy in a long thread.
In our run, "make them shorter and add a number" wrote 746 tokens but
read 61,731, because the entire conversation, both exports included,
came along. In a new conversation with a one-line brief, the same
request read 15,457. Copy work needs the campaign, the audience, and the
number to beat, not the raw data.
How to stop hitting Claude usage limits
Reduce the content and repeated work contributing to your usage:
- Check where you stand. On paid plans, Settings \> Usage shows
how much of the current five-hour session you've used, the time
left, and when your weekly limits reset. In Claude Code, /usage
shows your plan usage bars.
- Start fresh when the task changes. A new conversation drops the
history you'd otherwise re-send. In Claude Code, /clear does this
and /compact swaps the history for a summary.
- Lower the effort level for routine work. Higher effort uses more
tokens; Low and Medium can leave more allowance for subsequent
tasks.
- Treat the thinking tip with care. Anthropic's limits article
suggests turning extended thinking off, but its [model and effort
settings
guide](https://support.claude.com/en/articles/8664678-change-the-model-effort-and-thinking-settings)
says thinking can't be turned off in Claude on Haiku 5.5, Sonnet
- 5.5 or Opus 5.5. On those models, effort is your lever.
- Match the model to the job. In Claude Code, Opus "uses
meaningfully more of your quota" than Sonnet, and Haiku is the
cheapest option.
- Turn off connectors and tools a conversation doesn't need, and
ask Claude not to search the web when you don't need current
information.
- Put reusable reference files in a project. [Cached content counts
less against your
limits](https://support.claude.com/en/articles/9797557-usage-limit-best-practices)
while the cache remains active; after a long gap, the first message
counts that content in full again. Combine related questions into
one message when they use the same context.
- Turn on usage credits as a safety net. On Pro and Max, you
[enable usage
credits](https://support.claude.com/en/articles/12429409-manage-usage-credits-for-paid-claude-plans)
in Settings \> Usage and set a monthly spend limit. On Team and
seat-based Enterprise, an Owner or Primary Owner [enables
them](https://support.claude.com/en/articles/12005970-manage-usage-credits-for-team-and-seat-based-enterprise-plans)
in Organization settings \> Usage and sets org and per-user limits.
Either way, you keep working at standard API rates after a limit. On
Pro, Max, and Team, [usage
bundles](https://support.claude.com/en/articles/14246112-buy-usage-bundles)
prepay credits at up to 30% off.
For ad ops specifically
- Summarize before you paste. Export campaign-level totals for the
window you care about, not daily rows, so Claude reads a 12-row table
instead of 360 rows and doesn't have to add them up.
- One thread per task. Keep reporting in one conversation and start
copy or planning in a new one with a one-line summary.
- Use tools that return summaries, not raw rows. If you connect ad
accounts, prefer servers that return aggregated results and switch on
only what the task needs. Our MCP server, for example, lets you [load
only the tools a task
needs](https://docs.synterai.com/mcp/toolsets) through a
toolsets setting. If you're weighing options, compare [multi-channel
reporting
tools](https://synterai.com/blog/best-reporting-tools-multi-channel-ads)
on what they hand to Claude, not only on their dashboards.
When to stop running ad ops inside a chat window
Use conversation for analysis and decisions, such as investigating a CPA
change. For repeated pacing checks, negative-keyword sweeps, or weekly
reports, consider scheduled execution outside the chat. Running them
inside Claude still consumes plan usage, including [Claude Code
scheduled tasks](https://code.claude.com/docs/en/costs).
Recurring execution belongs in a system that runs it outside your Claude
session, and that is where Synter fits. [Synter's specialist
agents](https://synterai.com/features/ai-agents) for
creative, copy, analytics, budget, and targeting are on every plan per
our pricing, along with
negative keyword mining and search term harvesting, the autonomous CPA
optimizer, and budget pacing. What each can do in a given account
depends on the platform's API, your account permissions, and your
deployment scope. You can also schedule recurring checks from Claude
through [one MCP server for your ad
platforms](https://synterai.com/mcp): our
create_autonomous_schedule tool runs a skill on a cron schedule, and its
default is a read-only cross-platform pacing monitor that checks every 4
hours, alerts you in Slack, and never moves money.
Synter doesn't get you around Claude's limits, and it isn't free:
- Claude still counts what Claude reads. Setting up a schedule or
asking Claude about its results is a normal tool call, and the result
counts toward your Claude usage.
- Runs inside Synter are metered by us. Tool calls report to the
Developer Usage meter (reads are 2 credits, \$0.05), LLM tokens report
to Model Usage, and BYOK LLM calls skip Model Usage. Those meters sit
on your Synter bill, not your Claude plan.
- Writes go through approvals. Our background agents follow the
approval mode in Settings \> Autonomous. Its default, Ask before
acting, holds changes to your ad accounts while read-only work runs on
its own. Writes you request in the app, the API, or our hosted MCP
server follow the autonomy level instead: in production as of October
2026, the [default Suggest holds creating campaigns, enabling or
deleting things and sending creative for
approval](https://docs.synterai.com/mcp/approval-model),
while pausing or changing a budget runs immediately and is audited.
If you'd rather keep Claude as the operator, read our take on [running
Claude as an ads
agent](https://synterai.com/blog/claude-autonomous-agent-synter-mcp)
and keep its tools summary-first.
FAQ
How to avoid Claude usage limits?
Start a new conversation when the task changes, lower the effort level
for routine work, turn off connectors you don't need, and keep
reference files in a project, where cached content counts less against
your limits. For ad work, paste campaign-level summaries instead of raw
exports.
How many times can I use Claude?
There's no fixed number. Anthropic doesn't publish message counts for
any plan; it describes each tier as a multiple, such as Pro at least 5x
Free per session and Max 5x or 20x Pro. How far that goes depends on
conversation length, attached files, tools, model, and effort level.
What is the 5-hour limit in Claude?
It's the session limit. Free, Pro, Max, and Team plans have a usage
allowance that resets every five hours, and paid subscription plans add
weekly limits on top. On paid plans, Settings \> Usage shows how much of
the current session you've used and how long until it resets.
How to extend the usage limit on Claude?
Upgrade to a plan with a bigger multiple, or turn on usage credits so
you keep working at standard API rates after a limit, within a monthly
spend limit. Anthropic also hands out occasional [limit
resets](https://support.claude.com/en/articles/17007452-what-is-a-limit-reset)
to eligible plans, which you apply with "Reset for free" in Settings
\> Usage.
Do Claude Code and the Claude app share the same limit?
Yes. On Pro and Max, the [five-hour and weekly limits are shared across
Claude and Claude
Code](https://support.claude.com/en/articles/11145838-use-claude-code-with-your-pro-or-max-plan),
so a long terminal session leaves less for chat. One catch: if an
ANTHROPIC_API_KEY environment variable is set, Claude Code uses that key
and bills API usage instead of your subscription.
Conclusion
For paid subscription plans, session and weekly allowances are shared
across Claude surfaces and affected by conversation length, files,
tools, model, and effort. Check what your reporting workflow resends
before buying more capacity. Keep analysis in chat and consider moving
recurring execution outside that allowance.
To try summary-first reporting inside Claude, connect Synter's MCP
server and ask the cross-platform question you'd normally answer with
two CSVs. The page currently shows a \$100 ad credit voucher, code
MCP100.
discuss moving recurring pacing and reporting checks out of your Claude
sessions.