43  43. Claude Models, Plans, and Usage Limits

Which Claude model should you use? What does each plan include? And why do you sometimes see “limit reached”? This page answers those three questions (the products themselves are mapped in The Claude Product Family). Models and prices change often, so treat every number here as a snapshot from October 6, 2026.

43.1 The current models

As of Oct 16, 2026 · Official docs: Models overview · Choosing the right model

Claude is a family of models. The most capable models are slower and more expensive; the fastest ones are cheaper.

Model Anthropic’s description Speed Context window Good for
Claude Fable 5.1 “For demanding reasoning and long-horizon agentic work” Slower 1M tokens The hardest, longest tasks
Claude Opus 5.5 “For long-running agentic coding and knowledge work” Moderate 1M tokens Most work; Anthropic’s suggested starting point
Claude Sonnet 5.5 “The best combination of speed and intelligence” Fast 1M tokens Everyday coding and writing
Claude Haiku 5.5 “For high-volume, latency-sensitive tasks such as classification, extraction, and routing” Fastest 1M tokens Quick, simple, high-volume tasks
  • Names: the family name tells you the tier (from fastest to most capable: Haiku, Sonnet, Opus, Fable), and the number is the version. Fable 5.1 is Anthropic’s most capable model available to everyone.
  • Context window: how much text the model can work with at once. 1M (one million) tokens is roughly 555,000 words (see How LLMs Work).
  • Knowledge cutoff: all four current models have reliable knowledge through June 2026. A model knows about anything newer only if it searches the web or reads it in your files.
  • Model IDs and aliases: programs name a model by its ID, such as claude-opus-5-5. Claude Code also accepts short aliases such as opus, sonnet, haiku, and fable, which point to the current version (details in Context, Models, Effort, and Usage in Claude Code).
  • Claude Code’s default on paid plans is Opus 5.5. Fable is never the default; you choose it yourself, for example with /model fable.

43.2 Effort: how hard Claude thinks

As of Oct 6, 2026 · Official docs: Effort · Change the model, effort, and thinking settings · Model configuration

Effort is a dial for how much work Claude puts into each response. The levels, from least to most, are low, medium, high, xhigh (shown as “Extra high”), and max.

  • Higher effort means more careful reasoning and more checking of its own work, but it is slower and uses more tokens. That costs more on the API and uses up your plan’s limits faster.
  • Lower effort gives quicker answers and uses less, which is fine for routine work you will review anyway.
  • Each model has its own default. In Claude Code, Opus 5.5, Sonnet 5.5, and Haiku 5.5 start at medium. Some older models, such as Haiku 4.5, have no effort setting.
  • Anthropic’s advice: tuning effort is often a better lever than switching models.

In the Claude apps, effort is in the model menu next to the send button. In Claude Code, use /effort (details).

43.3 Plans and prices (a dated snapshot)

As of Oct 6, 2026 · Official docs: Claude pricing · What is the Pro plan? · What is the Max plan? · What is the Team plan?

Prices change: always check claude.com/pricing

These prices were copied from claude.com/pricing on October 6, 2026 (US dollars, before tax). Anthropic changes plans and prices regularly, and prices in the mobile app stores can differ. Check the official page before you buy anything.

Plan Price on Oct 6, 2026 Claude Code? Highlights
Free $0 No Chat on web, desktop, and mobile; web search, file creation, memory, connectors, artifacts; up to 5 projects; Sonnet and Haiku models only
Pro $20/month, or $17/month billed yearly ($200 up front) Yes At least 5× Free’s usage per 5-hour session; Opus; Claude Design, Slides, and Docs (beta); Research; Claude in Chrome and Microsoft 365
Max $100/month (Max 5x) or $200/month (Max 20x); monthly billing only Yes 5× or 20× Pro’s usage per session; Fable included (up to half of the weekly limit); higher output limits; early access to new features
Team Standard seat: $25/month ($20 billed yearly); Premium seat: $125/month ($100 yearly) Yes For 2 to 150 people; central billing and admin, single sign-on, Claude Tag (beta); a Premium seat gets 5× a Standard seat’s usage
Enterprise $20 per seat per month (billed yearly) plus all usage at API rates Yes For large organizations; admins set spending limits; advanced security and compliance
Education Not published Depends on your university A campus-wide plan that universities buy for students, faculty, and staff

For students: Anthropic says it offers no standing discounts on its paid plans (only occasional promotions). Your university may have a Claude for Education plan; check with your IT department.

43.4 Two ways to pay: subscription or API

As of Oct 16, 2026 · Official docs: API pricing · Why do I pay separately for the API? · Authentication

Subscription (Pro, Max, …) API (Claude Console)
You pay A fixed monthly price Per token used (“pay as you go”)
You get The Claude apps and Claude Code, up to your usage limits Access for your own programs, and for Claude Code if you sign in with an API key
If you use a lot You hit a limit and wait (or pay for extra) A bigger bill (you can set a monthly spending limit)
Managed at claude.ai, Settings platform.claude.com

The two are separate: a Pro plan does not include API usage, and API credit does not give you a Pro plan. API prices on October 16, 2026, in US dollars per million tokens:

Model Input Output
Fable 5.1 $10 $50
Opus 5.5 $4 $20
Sonnet 5.5 $2 $10
Haiku 5.5 $0.10 $0.50

For example, a request that sends Sonnet 5.5 100,000 tokens and gets 10,000 tokens back costs 0.1 × $2 + 0.01 × $10 = $0.30. Notice that output costs five times as much as input for every model in the table. (Haiku 5.5’s prices are for prompts up to 100,000 tokens; a longer prompt costs $0.50 and $2.50.) Batch requests (answered later, in bulk) are half price, and input that is sent again and again can be cached at a discount.

Watch out for ANTHROPIC_API_KEY

If the environment variable ANTHROPIC_API_KEY is set, Claude Code asks whether to use that key, and once you approve it, Claude Code uses the key instead of your subscription, billed at API rates. To go back to your subscription, remove the variable (for example unset ANTHROPIC_API_KEY, and delete it from your shell startup file), then check /status to see which sign-in is active. See Environment Variables.

43.5 Usage limits in plain English

As of Oct 6, 2026 · Official docs: How do usage and length limits work? · Manage usage credits · What is a limit reset?

A subscription does not give you a fixed number of messages. Instead you get an allowance of work, measured in tokens:

  • Two clocks. Every plan has a limit that resets on a rolling 5-hour window. Paid plans also have a weekly limit that resets at a fixed day and time for your account (shown in Settings > Usage on claude.ai).
  • One shared pool. claude.ai, the desktop and mobile apps, and Claude Code all draw from the same limits.
  • What uses it up faster: long conversations, big files, bigger models (Fable fastest), higher effort, tools such as Research and web search, multi-step agent tasks, and several sessions running in parallel.
  • Usage limit vs. length limit. The usage limit is about how much you use Claude over time. The length limit is the context window: how long a single conversation can get. They are different limits.

When you hit a usage limit, you can:

  1. Wait for the reset (the message tells you when).
  2. Upgrade to a bigger plan.
  3. Turn on usage credits (paid plans): pay as you go at API rates, with a monthly spending cap that you choose.
  4. Use a free limit reset if Anthropic has given you one (Settings > Usage, on the web or in Claude Desktop).

43.6 Checking your usage with /usage

As of Oct 6, 2026 · Official docs: Manage costs effectively · Commands

Inside a Claude Code session, type:

/usage
  • A Session block shows this session’s token counts and an estimated cost at API prices. It matters if you pay through the API (your real bill is in the Console); on a subscription, that dollar figure is not what you pay.
  • On a subscription, you also see usage bars for your plan’s limits (the current session and the week), each with its reset time, plus a breakdown of what is using up your limits: which skills, subagents, plugins, and MCP servers, and warnings about habits such as very long conversations. Press d or w to switch between the last day and the last week. The breakdown is approximate and counts only sessions on this computer, not your phone or claude.ai.
  • /cost and /stats are other names for /usage, and /usage-credits opens the usage-credit settings.

On claude.ai, the same information is in Settings > Usage. Ways to use fewer tokens (such as /clear between tasks) are covered in Context, Models, Effort, and Usage in Claude Code.

As of Oct 6, 2026 · Official docs: Claude Fable models on your plan · Fast mode · Code Review pricing

These are billed through usage credits (pay as you go), not your plan’s included usage:

  • Fable models on Pro and on Team Standard seats. (On Max and Team Premium seats, Fable can use up to half of the weekly limit; beyond that it needs usage credits.)
  • Fast mode in Claude Code (research preview): Opus up to 2.5× faster at a higher per-token price.
  • Code Review (research preview), the automatic pull-request reviewer on Team and Enterprise plans.

In interactive sessions, Claude Code asks for your consent before a Fable request bills usage credits.

As of Oct 16, 2026 · Official docs: Models overview · Model deprecations · Claude Platform release notes

  • Older models such as Fable 5, Opus 5, Sonnet 5, and Haiku 4.5 remain available on the API. Old models are eventually retired: for example, Sonnet 4.5 leaves the API on November 30, 2026. Programs that name a retired model stop working, so developers watch the deprecation announcements.
  • Claude Mythos 5.1 is offered by invitation only to vetted cybersecurity and biology research groups. Fable 5.1 is the same underlying model with extra safeguards.

43.7 Key takeaways

  • Start with the default model; switch models or effort when a task is unusually easy or hard.
  • A subscription has usage limits; the API bills per token. They are separate.
  • All your Claude use shares one pool of limits; check it with /usage.
  • Prices change: check claude.com/pricing.

Practice Questions

Easy

Question 1 · Matching

Match each model with its description.

Model Description
1. Claude Fable 5.1 A. The fastest current model, for high-volume tasks such as sorting messages into categories
2. Claude Opus 5.5 B. The most capable model available to everyone, for the hardest and longest tasks
3. Claude Sonnet 5.5 C. Anthropic’s suggested starting point for most work, including long agentic coding
4. Claude Haiku 5.5 D. “The best combination of speed and intelligence”

1 → B, 2 → C, 3 → D, 4 → A.

Question 2 · Multiple choice

Which current model has the largest context window?

  1. Claude Fable 5.1
  2. Claude Opus 5.5
  3. Claude Haiku 5.5
  4. They all have the same context window

d. All four current models (Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5) have a 1M-token context window. They differ in speed, cost, and how capable they are, not in how much text they can work with at once.

Question 3 · True or false

Your chats on claude.ai and your Claude Code sessions draw from separate usage limits.

False. claude.ai, the desktop and mobile apps, and Claude Code all draw from one shared pool of usage. Heavy chatting leaves less for Claude Code, and the other way around.

Question 4 · Short answer

What does the effort setting control? List the five effort levels from lowest to highest.

Effort controls how much work Claude puts into each response: how carefully it reasons and checks its work. Higher effort is more thorough but slower and uses more tokens. The levels are low, medium, high, xhigh (“Extra high”), and max.

Question 5 · Fill in the blank

Inside a Claude Code session, the command ________ shows your plan’s usage bars and what is using up your limits.

/usage (/cost and /stats are other names for the same command).

Question 6 · Multiple choice

How does the shorter of the two usage limits reset?

  1. Every hour, on the hour
  2. On a rolling 5-hour window
  3. Every day at midnight in your time zone
  4. Once a month, on your billing date

b. Every plan has a limit that resets on a rolling 5-hour window. Paid plans also have a weekly limit, which resets at a fixed day and time assigned to your account. Options a, c, and d are not how Claude’s limits work.

Medium

Question 7 · Scenario

Maria has a Pro plan. At 3 p.m., in the middle of a Claude Code session, she sees a message that she has reached her usage limit. List her options.

  1. Wait until the limit resets (the message says when).
  2. Upgrade to a bigger plan (Max 5x or Max 20x).
  3. Turn on usage credits to keep working, paying as she goes at API rates, ideally with a monthly spending cap.
  4. Use a free limit reset if she has been given one (Settings > Usage on the web or in Claude Desktop).

To stretch her limits afterwards she can use a smaller model or lower effort for routine work, and start fresh conversations between tasks.

Question 8 · Find the bug

Sam has a Pro plan. For a class project he added this line to his ~/.bashrc file:

export ANTHROPIC_API_KEY="sk-ant-..."

Now he sees API charges on his Console bill whenever he uses Claude Code. Explain what is happening and how he can fix it.

Because ANTHROPIC_API_KEY is set in every new terminal, Claude Code offers to use the key, and once Sam approved it, Claude Code uses the API key instead of his Pro subscription, so he is billed per token. Fix: delete the line from ~/.bashrc, run unset ANTHROPIC_API_KEY (or open a new terminal), start Claude Code again, and check /status to confirm that his claude.ai subscription is the active sign-in. If the class project needs the key, he can set it only in that project’s own environment instead of in ~/.bashrc.

Question 9 · Short answer

What is the difference between a usage limit and a length limit?

A usage limit controls how much you can use Claude over a period of time (the 5-hour and weekly allowances, shared across all apps and Claude Code). A length limit is the context window: how long one conversation can get before it no longer fits. Hitting a usage limit means waiting, upgrading, or paying for more; hitting a length limit means starting a new conversation (or letting Claude summarize the old one).

Question 10 · Multiple choice

Which of these will use up your plan’s limits fastest?

  1. A short question to Sonnet 5.5 at low effort
  2. A long Claude Code session on Fable 5.1 at max effort, with several sessions running in parallel
  3. Asking Haiku 5.5 to summarize a one-page PDF
  4. One web search on claude.ai

b. It combines everything that uses limits quickly: a long conversation, the most demanding model, the highest effort, and parallel sessions. Options a, c, and d are short, single tasks.

Question 11 · Scenario

Suggest a model (and, where it applies, an effort level) for each task, and give a reason.

  1. Rename a variable everywhere in one file.
  2. Plan a risky change that touches most of a large, unfamiliar codebase.
  3. A program that sorts 10,000 short customer emails into categories through the API, as cheaply as possible.
  1. A fast model such as Sonnet 5.5 (or the default model) at low effort: it is a simple, mechanical change that you will review anyway.
  2. Opus 5.5 at high or xhigh effort, or Fable 5.1 if that still falls short: the task needs deep reasoning, and mistakes are expensive.
  3. Haiku 5.5: the fastest and cheapest model, built for simple, high-volume work.

Other reasonable choices are fine if the reasoning matches the task’s difficulty, speed needs, and cost.

Question 12 · Short answer

What is a model alias in Claude Code? Give an example, one advantage, and one reason a developer might use a full model ID instead.

An alias is a short name that points to the current version of a model family, for example opus (which points to Opus 5.5 on October 6, 2026), sonnet, haiku, or fable. Advantage: you don’t have to remember version numbers, and you get the newest version automatically. A full ID such as claude-opus-5-5 pins one exact version, so behavior doesn’t change when the alias moves to a newer model.

Question 13 · Predict the output

Two students run /usage in Claude Code. One is signed in with a Pro subscription; the other uses a Console API key. Describe what each one sees.

  • Pro subscription: usage bars for the plan’s limits (current session and week) with reset times, plus a breakdown of what is using the limits (skills, subagents, plugins, MCP servers, and warnings about habits like very long conversations). Any session dollar estimate on the screen is not what a subscriber pays.
  • API key: the session’s token counts and an estimated dollar cost at API prices, with no plan usage bars. The real bill is in the Claude Console.

Hard

Question 14 · Compare

Should each of these use a subscription or the API? Explain.

  1. A student who wants to use Claude Code for coursework this semester.
  2. A startup building a customer-support chatbot for its website, used by thousands of customers.
  1. Subscription (for example Pro). It has a predictable monthly price, covers both the Claude apps and Claude Code, and the usage limits protect the student from a surprise bill.
  2. API. The chatbot is the startup’s own software calling Claude, which requires an API key; a personal subscription can’t be used inside a product. Costs grow with use, so the startup should pick the cheapest model that does the job well, set spending limits, and monitor usage in the Console.

Question 15 · Long answer

Anthropic says that “tuning effort is often a better lever than switching models.” Explain what this means and give an example.

Instead of jumping up to a more capable (slower, more expensive) model, or down to a faster, less capable one, you can often keep the same model and change how hard it thinks. Lower effort makes routine work faster and cheaper; higher effort gives more thorough reasoning on hard steps.

Example: if Opus 5.5 at its default medium effort misses an edge case in a tricky bug fix, first try high or xhigh on Opus 5.5 before switching to Fable 5.1. If simple edits feel slow, try low effort before switching to a faster model.

Question 16 · Scenario

Maria compares two numbers: the usage breakdown in /usage on her laptop, and Settings > Usage on claude.ai. Claude.ai says she has used much more of her weekly limit than her laptop’s breakdown suggests. She also uses Claude a lot on her phone. Explain the difference.

The breakdown in /usage is approximate and is calculated only from Claude Code sessions on that computer. It doesn’t include her phone, claude.ai, or other devices. Her plan’s limit is one shared pool across all of them, so claude.ai, which shows the whole account, reports more. (The plan usage bars come from her account, so for “how much is left,” trust the bars and Settings > Usage.)

Question 17 · Scenario

A blog post from June 2026 says, “Pro includes Fable at no extra cost.” But when Sam, who has a Pro plan, picks Fable 5.1 in October 2026, Claude Code says it requires usage credits. What is the most likely explanation, and how should Sam handle conflicting information like this in general?

Plans change. According to Anthropic’s Help Center, a promotion that let Pro users use the earlier Fable 5 model within their limits ended in July 2026, and Fable 5.1 was never included in Pro. On Pro, Fable models now run on usage credits; on Max (and Team Premium seats) they are part of the plan, up to half of the weekly limit.

In general: check the date of what you read, prefer the official pages (claude.com/pricing and support.claude.com), and trust what your own account shows today over an older blog post.

Question 18 · Long answer

A friend on a tight budget wants to use Claude Code all semester. Give advice about choosing a plan and about keeping usage and costs under control.

A good answer covers most of these points:

  • The Free plan does not include Claude Code, so a paid plan (usually Pro) or pay-per-use API billing is needed.
  • First check whether the university offers Claude for Education.
  • Pro has usage limits, which act like a built-in budget; upgrade to Max only if the limits regularly get in the way.
  • If turning on usage credits, set a monthly spending cap.
  • Stretch the limits: use the default or a smaller model and lower effort for routine tasks, /clear between unrelated tasks, and avoid running many sessions at once.
  • Make sure no ANTHROPIC_API_KEY is set by accident, so Claude Code uses the subscription.
  • Check /usage regularly, and check prices on claude.com/pricing, since they change.

Question 19 · Hands-on

Start a Claude Code session in any project folder and run:

/usage
/model
/effort status

Write down: (1) your usage bars and reset times, (2) the biggest contributors in the usage breakdown, if any, (3) your current model, and (4) your current effort level. Then close the /model picker without changing anything (press Esc).

Answers depend on your account. A good answer reports the session and weekly percentages with their reset times; names any skills, subagents, plugins, or MCP servers in the breakdown (a new user may have none); names the model (on a paid plan the default is Opus 5.5 unless you changed it); and gives the effort level (for Opus 5.5 the default is medium). If you signed in with an API key, you see token counts and an estimated cost instead of usage bars.

© 2025-2026 Y. Rosenthal. All rights reserved.