LLM Usage Limits 2026: ChatGPT vs. Claude vs. Gemini vs. Grok vs. Perplexity (Full Comparison)

🔄 Rewritten 16 September 2026. Every price, model, and limit below was re-checked against the official pricing pages, help centres, and announcements this week.

Two weeks into September, all four big chat assistants have a new top model or a new top tier. OpenAI shipped GPT-6 Astra. Anthropic shipped Claude Fable 5.1. Google shipped Gemini 3.8 Flash. And xAI, which now trades as SpaceXAI, added a $100 SuperGrok Plus plan between its $30 and $300 tiers.

Look closely at where those new models landed and a pattern shows up. On the $20 plans, the newest model is either missing from the main chat window or costs extra to use. ChatGPT Plus gets Astra in Work and Codex, but not in Chat. Claude Pro can use Fable 5.1 only by paying for usage credits. Gemini’s newest Flash starts at AI Pro, and its headline agent feature, Spark, is an Ultra perk.

So the question of which plan to buy has changed shape. A year ago you picked a plan by how many messages you needed. Today you pick by two things: how much work you do (the limit system) and which model you need for it (the model gate). This guide covers both for all five platforms.

What’s Changed Since 1 September 2026

OpenAI launched GPT-6 Astra (September 3). OpenAI calls it its most capable model, with big gains in computer use, coding, and professional documents. In ChatGPT it arrives in two forms. GPT-6 Pro, powered by Astra, sits in the Chat picker on Pro ($100 and $200), Business, and Enterprise. Plus gets Astra in ChatGPT Work and Codex only. Pro-model messages are capped: 50 a week on Pro $100 (shared with GPT-5.6 Sol Pro) and 200 a week on Pro $200. (Sources: OpenAI — GPT-6 Astra: A new generation of intelligence; OpenAI Help Center — GPT-5.6 and GPT-6 Pro in ChatGPT.)

Anthropic launched Claude Fable 5.1 (September 1). It’s Anthropic’s strongest generally available model, and on typical token-billed workloads it’s about 25% cheaper to run than Fable 5 because cache reads now cost 75% less. On consumer plans it follows the Fable 5 rules. Max can spend up to half of its weekly limit on Fable models. Pro pays for Fable with usage credits from the first message. Free doesn’t get Fable at all. (Sources: Anthropic — Introducing Claude Fable 5.1 and Claude Mythos 5.1; Claude Help Center — Claude Fable models on your plan.)

Claude Code’s weekly limits dropped to their new permanent level (September 14). The +50% summer promotion is over. Standard weekly limits in Claude Code are now permanently 25% higher than they were before the promotion. Anthropic’s own maths: that’s about 17% less than the boosted limits most people had until last week. It covers Pro, Max, Team, and seat-based Enterprise. (Source: Claude Code announcement, August 29, 2026.)

Google launched Gemini 3.8 Flash (September 2), its third Flash release in six weeks. In the consumer app it’s for Google AI Pro and Ultra subscribers, and it also reaches AI Mode in Search and Gemini in Sheets. Free users stay on 3.6 Flash. Google also put a native Gemini app on Windows on September 10. (Sources: Google — Introducing Gemini 3.8 Flash and 3.8 Flash Cyber; Gemini Apps release notes.)

Gemini limits can now be extended with AI credits. Gemini’s plan page now says you can buy AI credits once you run through your compute allowance. Until now, a Gemini user who hit the weekly ceiling could only wait or upgrade. (Source: gemini.google/subscriptions.)

Grok got a $100 tier. SuperGrok Plus sits between SuperGrok ($30) and SuperGrok Heavy ($300), with much higher usage across Chat, Imagine, Voice, and Build, 1080p video, and priority access at peak times. Grok 4.6 is still the flagship. Grok 4.7 was publicly targeted for September 12 but isn’t on SpaceXAI’s pricing page as of today. (Source: x.ai/pricing.)

ChatGPT tidied up how it thinks and talks.

• September 14: Plus and Pro stopped switching automatically from Instant to Thinking. If you want deeper reasoning, pick it in the model picker.
• September 9: Voice got simpler daily limits: Go 3 hours (GPT-Live-1 mini), Plus 3 hours, Pro $100 15 hours, and Pro $200 unlimited.
• September 11: OpenAI announced it plans to retire custom GPTs and move them to plugins. Dates vary by plan.

(Source: ChatGPT release notes.)

Perplexity’s plans stayed put. Pro is still $20 a month and Max is still $200. The model lineup now shows Gemini 3.8 Flash alongside GPT-5.6 Terra, Claude Sonnet 5, and Grok 4.6. (Source: perplexity.ai/pro.)

Quick Comparison: All Five Platforms at a Glance

Where the Newest Model Actually Lives on Each Plan

This is the table most comparison pages skip, and it’s the one that decides whether an upgrade gets you what you’re paying for.

Grok is the exception. SpaceXAI charges for more usage, and every paid tier gets the same flagship. Everywhere else, the newest model sits behind its own gate, and that gate is often higher than the plan most people are on.

ChatGPT (OpenAI)

Plans: Free · Go $8/month · Plus $20/month · Pro $100/month · Pro $200/month.

ChatGPT now has two separate limit systems, and you need to know which one you’re in.

1. Chat: the main ChatGPT window. Free and Go users run on GPT-5.6 Luna with unlimited everyday text chats. OpenAI still applies abuse safeguards. File uploads, image generation, voice, data analysis, and other tools have their own limits. Paid plans run on GPT-5.6 Sol, and a picker sets how hard it thinks: Instant, Medium, High, Extra High, and Pro.

2. Work and Codex: where agents and longer tasks run. These share one allowance that uses a five-hour window plus a weekly limit, and you need room left in both. OpenAI publishes estimated ranges instead of fixed caps. For GPT-6 Astra, Plus gets roughly 5–45 messages per five hours, Pro $100 gets 25–225, and Pro $200 gets 100–900. GPT-5.6 Sol stretches about twice as far on each plan. Check Settings → Usage to see where you stand.

What each plan gets:

• Free — GPT-5.6 Luna, unlimited text, and a Think button (still Luna) for harder questions. Ads appear in some regions.
• Go ($8) — the same unlimited Luna text with more room for uploads, images, and other tools. Voice is capped at 3 hours a day on GPT-Live-1 mini.
• Plus ($20) — GPT-5.6 Sol at Instant, Medium, and High. GPT-6 Astra in Work and Codex, but no Astra in Chat and no Extra High or Pro options. Voice is 3 hours a day.
• Pro ($100) — every Sol level including Extra High and Pro, plus GPT-6 Pro in Chat at 50 messages a week, shared with GPT-5.6 Sol Pro. Work and Codex get about 5× Plus. Voice is 15 hours a day.
• Pro ($200) — 200 GPT-6 Pro messages a week, plus 170 Sol Pro messages a day (both together capped at 200 a day). Work and Codex get about 20× Plus. Voice is unlimited.

When you hit a wall: in Chat, ChatGPT may carry on with another available model. On Pro $200, running out of GPT-6 Pro drops you to GPT-5.6 Thinking at Medium. In Work and Codex, eligible Plus and Pro accounts can buy an instant reset, and credits can pay for extra usage. A reset refreshes your allowance but doesn’t raise your plan’s limits. OpenAI Support won’t reset limits for you.

Changes to plan around:

• As of September 14, Plus and Pro no longer switch automatically from Instant to Thinking on hard questions. If your answers suddenly feel shallower, raise the level in the picker yourself.
• Custom GPTs are being retired in favour of plugins. The dates vary by plan, so list the GPTs you rely on now while you still can.
• Box, Dropbox, and SharePoint joined Google Drive in ChatGPT’s Library (Go and up), so you can pull files in without uploading them again.

Watch out for: GPT-6 Astra usage comes out of the same Work and Codex allowance as everything else there, and Astra uses it faster than Sol. OpenAI’s own guidance is useful here: Astra at Low effort can beat Sol at High, so try a lower effort setting before you turn it up.

Saving your work out of ChatGPT is its own small adventure. The ChatGPT export guide covers chats, files, and images.

Claude (Anthropic)

Plans: Free · Pro $17/month (annual) or $20 (monthly) · Max 5× $100/month · Max 20× $200/month.

The model lineup, and who gets what:

• Haiku 4.5 — fast and cheap. Every plan.
• Sonnet 5 — the everyday default. Every plan, up to a 1M context window.
• Opus 5 — the flagship for long agentic work. Pro and Max only.
• Fable 5 and Fable 5.1 — the top-end models. Paid plans only, and each paid plan handles them differently.

Fable, plan by plan. Anthropic’s help centre treats Fable 5 and Fable 5.1 the same way:

• Max (and premium Team seats): Fable is part of the plan. You can spend up to 50% of your weekly limit on Fable. That 50% is a ceiling inside your normal limit, so it doesn’t add any capacity. After that, pay with usage credits or switch models.
• Pro (and standard Team seats): Fable isn’t included in your limits. It runs on pay-as-you-go usage credits from the first message.
• Free: no Fable.

The promotion that let Pro users run Fable 5 inside their plan ended on July 19, and Fable 5.1 was never part of it. If you remember Fable being “included” on Pro, that’s why.

Fable 5.1 is cheaper to pay for. Usage credits are billed at standard API rates, and Anthropic cut Fable 5.1’s cache-read price by 75%. That makes typical workloads about 25% cheaper than Fable 5, and up to about 45% cheaper for long agentic sessions. On Claude.ai and in Cowork it defaults to Medium effort. In Claude Code it defaults to High.

How the limits work: every plan has a rolling five-hour session limit, and paid plans add a weekly limit on top. Chat, desktop, mobile, and Claude Code all draw from one pool. Anthropic’s ratios are all per five-hour session: Pro gives at least 5× Free, and Max gives 5× or 20× Pro. There’s no fixed message count, because longer chats, bigger models, and heavier features all use more. When you hit a limit, you can wait, upgrade, or turn on usage credits. Settings → Usage shows where you are.

Pick your tier by the model you need, then by how hard you’ll use it:

• Pro ($17–$20) — steady daily use on Sonnet 5 and Opus 5. Fable is available if you’re willing to pay for credits.
• Max 5× ($100) — long sessions, plus Fable 5.1 inside your plan. For most people this is the cheapest way to make Fable an everyday model.
• Max 20× ($200) — all-day use, several parallel threads, or Claude Code as your main tool.

If you’ve already hit a Claude limit mid-task, here’s what actually happens and how to keep working. And if Claude Code is the main reason you pay, read the Codex vs Claude Code comparison before you commit to a tier.

Gemini (Google)

Plans: Free · Google AI Plus $4.99/month · Google AI Pro $19.99/month · Google AI Ultra $99.99 (5× Pro) or $199.99 (20× Pro)/month.

Google has the widest price ladder here, and every step bundles storage and Google-app extras. That makes it easy to compare badly. Here’s what each tier does inside the Gemini app:

• Free — Gemini 3.6 Flash, some access to 3.1 Pro, image generation, Deep Research, Gemini Live, Canvas, and Gems. 15 GB of storage.
• AI Plus ($4.99) — 2× Free usage, plus video generation, Daily Brief, 200 Google Flow credits, and 400 GB of storage.
• AI Pro ($19.99) — 4× Free usage, Gemini 3.8 Flash, 1,000 Flow credits, higher limits for the Jules coding agent, and 5 TB of storage.
• AI Ultra ($99.99 / $199.99) — 5× or 20× AI Pro usage, Deep Think, first access to Gemini Spark (in select countries), Project Genie, 10,000 or 25,000 Flow credits, and 20 TB or more of storage.

The models: Gemini 3.1 Pro is still the large flagship. Gemini 3.8 Flash, released September 2, is the new workhorse and Google’s best model for coding and long, multi-step tasks. It works harder on complex prompts, and Google says it can use more tokens to get there. Three Flash releases in six weeks means whatever you read about “the Flash model” in August is probably out of date.

How the limits work: Gemini measures compute. Your usage depends on how complex the prompt is, which features you use, and how long the chat runs. The allowance refreshes every five hours until you reach a weekly limit. The new part is that you can now buy AI credits to keep going once you’re out, so Gemini works like Claude and Grok here.

Also new:

• A native Windows app (September 10). Press Alt+Space to open it from anywhere, and it connects to Gmail and Drive.
• Students: eligible US college students can get a year of Google AI Pro free, or AI Plus in 140+ countries. The offer runs through December 31, 2026.

Watch out for: AI Plus doesn’t include 3.8 Flash. If the newest Flash model is the reason you’re upgrading, the cheapest plan that has it is AI Pro at $19.99.

Two of Gemini’s strongest tools have their own walkthroughs: Gemini Deep Research, and the Gemini 3 Flash vs Pro guide if you’re deciding which model to use.

Grok (SpaceXAI)

Plans: Free · SuperGrok Lite · SuperGrok $30/month · SuperGrok Plus $100/month · SuperGrok Heavy $300/month.

First, the name. xAI now operates as SpaceXAI, so the pricing page, the docs, and the footer all say SpaceXAI. Grok is still called Grok.

Grok 4.6 remains the flagship, with a 500,000-token context window. Grok 4.7 has been promised publicly and has already missed at least one target date. It isn’t on the pricing page as of September 16, so don’t buy a plan expecting it.

• Free — real-time web and X search, voice, and connectors, within limits SpaceXAI calls “generous” but doesn’t publish.
• SuperGrok Lite — the entry paid tier, aimed mainly at Imagine (images and video).
• SuperGrok ($30) — Grok 4.6, Grok Bot, connectors, Expert mode, image and video generation, and higher limits everywhere.
• SuperGrok Plus ($100) — everything in SuperGrok, plus much higher usage across Chat, Imagine, Voice, and Build, 1080p video, faster replies, and priority access at peak times.
• SuperGrok Heavy ($300) — the largest weekly allowance, for the heaviest users.

How the limits work: every paid plan gets one shared weekly pool that covers Chat, Imagine, Voice, and Build. Spend it however you like, and check your reset time under Settings → Usage. When the pool runs out, paid features pause, but you drop back to free-tier Chat and Voice, which have their own separate limits. You don’t get locked out completely.

If you run dry: buy Extra Usage Credits from $5, or turn on automatic top-ups. SpaceXAI is upfront about the trade-off: credits cost more per action than your plan’s included usage. If you top up every week, you’re paying more than the next tier would cost.

Watch out for: the shared pool makes media expensive. One video generation uses far more of your week than a chat does. Before SuperGrok Plus launched, heavy Imagine users had to choose between a $30 plan that ran out and a $300 one. Plus at $100 is now the obvious step up for them.

Perplexity

Plans: Free · Pro $20/month ($17/month billed annually) · Max $200/month ($167/month billed annually).

Perplexity is built around search with citations, so its limits work differently. Ordinary searches are free and unmetered on every tier. You pay for deeper research, model choice, and Computer, Perplexity’s agent that carries out multi-step tasks.

• Free — unlimited standard searches with citations, a small daily number of Pro Searches, and limited Research.
• Pro ($20) — Deep Research, a choice of models, more usage and memory, expanded Computer access, and a batch of bonus credits (currently 4,000, plus a limited-time $40 in Computer credits).
• Max ($200) — the highest limits, 10,000 Computer credits a month plus bonus credits, the top frontier models, Model Council, and early access to new features.

The credits system: only Computer spends credits. Asking a question never does. Consumer Max accounts start with a default $200 spending cap on credit charges, and you can raise it to $5,000 in settings. If you only search and research, you’ll hardly notice credits exist. If you hand Computer long jobs, credits become the real limit.

The models: the Pro lineup currently includes GPT-5.6 Terra, Gemini 3.8 Flash, Claude Sonnet 5, Kimi K3, GLM 5.3, and Grok 4.6. Max adds the frontier tier on top. Perplexity doesn’t have its own flagship model. It licenses other companies’ models and adds them soon after release, which is why Gemini 3.8 Flash was already there within two weeks.

Recent additions: you can now start Computer tasks from email, and Computer can run sub-tasks on GPT-5.6 Terra and Luna (August 24).

How Usage Limits Work Now

“40 messages every three hours” is gone. Here’s what replaced it, and it’s now fairly consistent across the platforms.

The five-hour plus weekly model is now the standard. Claude, Gemini, ChatGPT’s Work and Codex, and (on a weekly basis) Grok all measure compute: how much work each request costs. A quick question costs little. A long agent run, a video, or a high-effort reasoning pass costs a lot. Usage refills in rolling windows up to a weekly ceiling. The practical takeaway: you can’t compare plans by message counts any more, because the same plan feels generous to someone asking lots of short questions and tight to someone running two long coding sessions a day.

Paying to keep going is now standard too. Every platform now has a way to pay past your limit:

Keep one rule in mind with all of these: pay-as-you-go always costs more per action than your plan’s included usage. Grok says so in its docs, and Claude’s credits bill at full API rates. Topping up now and then is fine. Topping up every week means you’re on the wrong plan.

Text is cheap, tools and agents are expensive. ChatGPT made this obvious by giving free users unlimited text while keeping limits on files, images, voice, and tools. Other platforms price the same way without saying it so plainly. Upgrading because you run out of plain chat makes less sense every month. Running out of uploads, agent time, or top-model access are the real reasons to pay.

The same model can cost very different amounts depending on effort. ChatGPT, Claude, and Gemini all let you, or the app, choose how hard the model thinks. Higher effort uses more of your allowance and doesn’t always give a better answer. Before assuming you need a bigger plan, try a newer model at lower effort.

Promotions end. Claude Code’s +50% summer boost ended as a permanent +25% over the old baseline, which is still a cut from what people had during the promotion. ChatGPT’s launch-week resets for Astra were one-offs. When a platform raises your limits “for now”, look up the baseline before you plan around the higher number.

Which Plan Should You Buy?

Start by working out which limit you’re actually hitting. This is the decision rule for everything else. There are three possibilities, each with a different fix:

1. Volume: you run out of messages or session time. Buy more headroom on the platform you already use, or add a second platform’s free tier.
2. Tools: uploads, images, voice, or agent time run out. These need a paid tier, and on ChatGPT that’s the main thing Go and Plus give you over Free.
3. Model: you need the newest model for the work. Look at the “Where the newest model lives” table above, because that’s often the $100 plan or credits.

For most people, the $20 plan is still the right choice. ChatGPT Plus, Claude Pro, Gemini AI Pro, and Perplexity Pro all give you strong models, real research tools, and enough room for daily professional work. Grok’s equivalent is SuperGrok at $30. Choose the one that fits the tools and apps you already use. Benchmark leads don’t last: every one of these platforms has shipped a new top model this quarter, and it’ll happen again.

Upgrade to about $100 only for a specific model or heavy agent work. ChatGPT Pro $100 is the cheapest way to get GPT-6 Pro in the chat window. Claude Max 5× is the cheapest way to use Fable 5.1 without paying per message. SuperGrok Plus suits heavy Imagine and Build users. Gemini Ultra $99.99 is mainly for Deep Think, Spark, and the video credits.

Plans under $10 are a narrow buy. Gemini AI Plus ($4.99) doesn’t include 3.8 Flash. ChatGPT Go ($8) gives you the same unlimited text as Free, plus more room for tools. Both are fine if that extra room is exactly what you need, but neither gets you a better model.

Plan for credits if you use agents. Claude Code, ChatGPT’s Work and Codex, Perplexity Computer, and Grok Build all draw heavily on your allowance. With serious agent work, the subscription is only part of what you’ll spend.

Keep a backup. Pay for one primary platform and keep free accounts on the others. When you hit a weekly limit, or a model gets pulled or rationed, you can switch instead of waiting. A second model is also useful for checking the first one’s work. Here’s how to set that up.

Key Takeaways

• Every chat platform released a new top model or tier this month: GPT-6 Astra (Sept 3), Claude Fable 5.1 (Sept 1), Gemini 3.8 Flash (Sept 2), and SuperGrok Plus at $100.
• The $20 plans don’t get the newest model in the main chat window for free. Plus gets Astra only in Work and Codex, Claude Pro pays for Fable with credits, and Gemini’s newest Flash starts at AI Pro.
• Four platforms now have a plan at about $100, and that’s the tier where the newest models are usable day to day.
• Claude Code’s weekly limits fell about 17% on September 14, as the summer promotion turned into a permanent +25% over the old baseline.
• Every platform now lets you pay past your limit. Gemini added AI credits this month. Pay-as-you-go always costs more per action than a plan’s included usage.
• Work out which limit you’re hitting (volume, tools, or model) before you upgrade. Each one has a different fix.

Keep going:

• Claude usage limit reached? What actually happens, and how to keep working
• Use a second AI model to review the first
• Codex vs Claude Code: which should you use?
• Gemini 3 Flash vs Pro: which model to use
• AI in 2026: a quick peek for people who don’t want to become developers
• Export guides: ChatGPT · Claude · Gemini

Last updated: September 16, 2026. Sources: OpenAI — GPT-6 Astra: A new generation of intelligence (September 3, 2026), GPT-5.6 and GPT-6 Pro in ChatGPT, Managing usage with GPT-6 Astra in Work and Codex, and the ChatGPT release notes; Anthropic — Introducing Claude Fable 5.1 and Claude Mythos 5.1 (September 1, 2026), claude.com/pricing, Claude Fable models on your plan, and the Claude Code announcement of August 29, 2026; Google — Introducing Gemini 3.8 Flash and 3.8 Flash Cyber (September 2, 2026), gemini.google/subscriptions, and the Gemini Apps release notes; SpaceXAI — x.ai/pricing and the Grok FAQ; Perplexity — perplexity.ai/pro and the Perplexity changelog.