AI Usage Limits Compared (Oct 2026): ChatGPT, Claude, Gemini

Update (October 11, 2026): Three model changes since the September numbers below. ChatGPT: on October 7 OpenAI moved the regular Chat tab to GPT-6 Sol on Plus, Pro, Business, and Enterprise, and to GPT-6 Luna on Free and Go (rolling out from October 8). The models in ChatGPT Work and Codex did not change, so the Codex ranges below still apply, and OpenAI still publishes no fixed message cap for ordinary chat. Claude: Sonnet 5.5 (September 28) replaced Sonnet 5 as the current Sonnet, and Haiku 5.5 (October 7) replaced Haiku 4.5 as the current Haiku. Anthropic has not said which Sonnet or Haiku version each plan runs and does not publish message counts, so the Claude figures below are unchanged.

TL;DR. Nobody can promise one universal message count — all three major AI vendors meter compute, and the numbers flex with model and task complexity. OpenAI now publishes useful Codex ranges: on Plus, approximately 10–100 Sol, 25–200 Terra, or 250–2,000 Luna local messages per five hours; Pro multiplies those ranges by 5× or 20×, and since September 2026 paid Codex plans can also run GPT-6 Astra at roughly half the Sol message count (OpenAI's approximate figure). OpenAI still does not publish one dependable cap for ordinary ChatGPT chat. Claude Pro is commonly estimated around 45 messages/5h plus a weekly cap, and Anthropic raised paid plans' five-hour limits on September 22, 2026 (new numbers unpublished); Gemini uses compute-based allowances refreshing every five hours. Image generation has separate limits, covered below.

"How many messages do I actually get?" is the most reasonable question in AI subscriptions — and the one no pricing page will answer plainly. That's not entirely evasion: all three vendors now meter compute rather than counting messages, so the honest answer genuinely is "it depends on what your messages look like." But "it depends" plus real numbers beats "it depends" alone. Here are the real numbers as of September 2026, with each labeled as official or user-reported.

One rule before the tables: a "message" is not a unit of anything. A two-line question costs a sliver of allowance; a 200-page PDF upload with a long back-and-forth costs orders of magnitude more. Every number below assumes typical, moderate-length use.

How the metering actually works

New to the jargon? The AI glossary explains tokens, context windows, and usage caps in plain English.

All three platforms share the same architecture, with different vocabulary:

• Rolling windows, not daily resets. Your allowance replenishes on a rolling basis. OpenAI's current Codex estimates use five-hour windows; other ChatGPT features can use different windows. Claude and Gemini also commonly use five-hour windows. There is no universal midnight reset, and unused allowance does not bank.
• Weekly caps on top. Claude and Gemini add an explicit weekly ceiling above the rolling window; ChatGPT applies weekly caps to specific heavy features (reasoning modes) rather than all chat.
• Compute-weighted drawdown. Reasoning/thinking modes, deep research, large attachments, long conversation histories, and image or video generation all consume allowance faster than plain short messages — on every platform.
• Dynamic adjustment. All three vendors explicitly reserve the right to flex limits with demand, and all three exercised it in 2026. Any specific number has a shelf life.

ChatGPT: Free, Go, Plus, Pro

OpenAI publishes plan features but no single fixed cap for ordinary ChatGPT chat. It now publishes approximate model-specific ranges for Codex local messages, which are more useful than the older, user-reported "160 messages" figure because they show how strongly model choice changes consumption. (Full plan breakdown here.)

Plan	Ordinary ChatGPT chat	Published Codex local-message estimate
Free ($0)	"Unlimited text chats with GPT-5.6 Luna" (subject to abuse guardrails); "limited messages with uploads"	No published GPT-5.6 Codex range on the pricing table
Go ($8/mo)	Same unlimited Luna text chats; "more messages with tools," "more uploads," "more image creation" than Free	No published GPT-5.6 Codex range on the pricing table
Plus ($20/mo)	Plan-limited; no single public count	Sol 10–100 / Terra 25–200 / Luna 250–2,000 per 5h; GPT-6 Astra (limited allowance) at roughly half the Sol count
Pro 5× ($100/mo)	Higher limits; core chat subject to abuse guardrails; GPT-6 Astra in chat (about 50 "GPT-6 Pro" messages a week)	Sol 50–500 / Terra 125–1,000 / Luna 1,250–10,000 per 5h; Astra at roughly half the Sol count
Pro 200 ($200/mo; reopened to new sign-ups Sept 29, 2026)	Higher limits than Pro 100; GPT-6 Astra in chat. New subscribers get a lower allowance than before; subscribers active at the cutoff keep the old one through Oct 29, 2026	The Sol 200–2,000 / Terra 500–4,000 / Luna 5,000–40,000 per 5h ranges were the old 20× allowance; OpenAI has not published the new one
Pro 500 ($500/mo; new Sept 29, 2026)	"The highest included usage" of the three Pro plans; the only Pro tier with Ultrafast	Not published

These are broad estimates, not entitlements. OpenAI says model choice, context, reasoning, tool use, retrieval, and caching all affect usage, so two prompts that look similar can consume different amounts. The practical lesson is now explicit: save GPT-6 Astra (available since September 3, 2026) for the hardest end-to-end work, use Sol for hard work, Terra for everyday production tasks, and Luna for fast high-volume work when you want the same subscription allowance to stretch much further. The Astra figure — roughly half as many messages as Sol per five-hour window — is OpenAI's approximate number, not a published range.

ChatGPT Go vs Plus limits

The $8-vs-$20 question comes down to what is capped, not a message count — OpenAI's pricing page (checked September 6, 2026) describes both tiers relatively, never numerically:

	Go ($8/mo)	Plus ($20/mo)
Text chat	Unlimited text chats with GPT-5.6 Luna (abuse guardrails); chat moves to GPT-6 Luna from October 8	Same, plus "advanced reasoning models with GPT-5.6" — Sol; chat runs GPT-6 Sol since October 7 (GPT-6 Astra in chat starts at Pro; Plus gets it in Work and Codex only)
Messages & uploads	"More messages with tools," "more uploads" than Free	"Expanded messages and uploads"
Images	"More image creation" (observed: a couple dozen a day)	"More complex and accurate image creation" (observed: ~40–50 per 3-hour window)
Codex	Limited Codex access, no Sites	"Expanded Codex usage," Sites included
Ads	May include ads	No ads

So Go buys room on the Free model; Plus buys the better model and the agents. If you regularly hit a wall on Go, the wall is usually Luna, not the message count. The full tier-by-tier breakdown is in ChatGPT pricing.

Claude: Free, Pro, Max

Anthropic is the most explicit that it doesn't publish message counts — limits are token-based, and your mileage varies with conversation length and attachments. Two limits run simultaneously: a 5-hour rolling window and a weekly cap (which resets 7 days after your first message). You can see your own remaining allowance under Settings → Usage — Claude is the only one of the three that shows you a live meter. (For what each tier costs and which ones include Cowork, see Claude pricing.)

Plan	Practical limits (Sept 2026)	Status
Free ($0)	Roughly 15–40 messages per 5-hour window, varying with length, demand, and attachments	User-reported
Pro ($17–$20/mo)	~45 messages per 5-hour window for average conversations (5-hour caps were doubled in May 2026), plus the weekly cap	User-reported
Max 5x ($100/mo)	~225 messages per 5-hour window	User-reported
Max 20x ($200/mo)	~900 messages per 5-hour window	User-reported

Updated September 22, 2026: With the Claude Opus 5.5 launch, Anthropic raised five-hour usage limits on Pro, Max, Team, and seat-based Enterprise plans, and gave subscribers a one-time rate-limit reset they can save and use whenever they choose. Anthropic did not publish the new numbers, so the user-reported figures above predate the increase and are likely low.

Three Claude-specific wrinkles worth knowing in September 2026:

• A savable reset. The one-time rate-limit reset from September 22 is worth holding for a deadline day: use it when you hit a limit at the worst moment, rather than the first time you hit one.

• Fable 5.1 has its own budget. Anthropic's most capable generally available model — Fable 5.1 since September 1, 2026 — draws from only 50% of weekly limits on Max and Team Premium, and Pro doesn't include it at all: Pro users can run it only on pay-as-you-go usage credits, and Free gets none. If you're on Max and use Fable 5.1, it consumes your allowance at a different rate than Sonnet. What changed in 5.1 · how the split started.

• Model choice is your throttle. Running everyday tasks on Sonnet (Sonnet 5.5 is current since September 28) rather than the heavier models stretches the same plan dramatically — Sonnet matches the bigger models on most routine professional work anyway.

Gemini: Free, AI Plus, AI Pro, AI Ultra

Update (October 2026): Google is also changing which models each plan can use. Starting October 9, 2026, the free plan uses Auto model selection (most prompts go to Flash-Lite; Google may route prompts that need deeper reasoning to Flash or Pro, and free users can't pick either themselves); AI Plus keeps Flash but loses Pro on a date Google emails subscribers; AI Pro and Ultra keep all models. Google's help page first said AI Pro also gains Deep Think, but as of October 6 its US version listed Deep Think for AI Ultra only, and as of October 13 it no longer mentions Deep Think. Every plan gets low/medium/high effort levels, and higher effort uses more of your allowance. What changes on each Gemini plan.

Google made the metering shift explicit at I/O 2026: the Gemini app moved from per-model daily prompt caps to compute-based allowances — every request draws down an allowance based on complexity, model, and chat length, refreshing every 5 hours until a weekly ceiling.

Plan	Allowance (Sept 2026)	Status
Free ($0)	Base allowance; heavy features throttle first	Official (unquantified)
AI Plus ($4.99/mo)	~2× the free allowance	Official multiplier
AI Pro ($19.99/mo)	~4× the free allowance; Gemini 3.1 Pro plus the new Gemini 3.8 Flash available	Official multiplier
AI Ultra ($99.99–$199.99/mo)	5×–20× beyond Pro depending on tier; Deep Think (Google's wording on AI Pro access changed in October 2026); includes Gemini Spark	Official multiplier

Google publishes multipliers rather than absolute numbers, which is honest about the mechanism but useless for planning — the practical guidance from heavy users: normal chat rarely hits limits on any paid tier; Deep Research, Deep Think, and image/video generation are what burn the allowance, on Free especially.

Image generation limits — the question everyone actually asks

A reader recently asked us directly: "How many image generations do you get on the Go plan vs Pro?" Here's the honest answer for all three vendors, as of September 2026:

ChatGPT. OpenAI does not publish fixed image counts for any plan. The observed reality: Free gets roughly 2–3 images per day. Go is marketed as ~10× Free's image creation — call it 20–30/day in practice, flexing with demand. Plus users commonly report ~40–50 images per rolling 3-hour window. Pro is "unlimited and faster image creation, subject to abuse guardrails" — OpenAI's own wording, and the only tier where you can stop counting. So Go-vs-Pro specifically: it's roughly "a couple dozen a day, dynamically throttled" versus "don't think about it," with a $92–192/month gap between them — most people who feel Go's image ceiling should try Plus first.

Gemini is the only vendor publishing actual numbers, via its help pages: Free up to ~20 Nano Banana 2 images/day (with only ~3/day on the top image model), AI Plus 50/day, AI Pro 100/day, AI Ultra 1,000/day — resetting daily at midnight Pacific, and explicitly subject to change under demand. At the $20 tier, Gemini's 100/day is the most generous published image allowance in the market.

Claude doesn't do native image generation at all — no plan includes it. Claude's visual output is diagrams, charts, and documents via Artifacts, not photorealistic image synthesis. If image generation matters to your work, that's a genuine reason to run Claude alongside ChatGPT or Gemini rather than instead of them — a common pairing among professionals comparing the three.

So which plan do you actually need?

Light user (a few questions a day): every Free tier covers you. You'll feel ChatGPT Free's model fallback first; Claude and Gemini Free degrade more gracefully for chat.

Daily professional use, mostly text: any $20 tier — ChatGPT Plus, Claude Pro, Gemini AI Pro — is sized for a normal workday. Choose on which tool fits your work, not on limit arithmetic; at typical usage you won't hit any of them often.

Heavy agentic use, long documents, all-day sessions: this is what the $100–200 tiers (ChatGPT Pro, Claude Max, Google AI Ultra) are actually for. ChatGPT's $200 Pro 200 tier reopened to new sign-ups on September 29, 2026 with a smaller allowance than before, and a $500 Pro 500 tier now sits above it, so the top personal rungs are ChatGPT Pro 500, Claude Max 20×, and Google AI Ultra 20×. Before paying 5–10× more, spend one week economizing — shorter conversations, cheaper models for routine tasks, heavy modes only when needed. Many "I keep hitting limits" cases are workflow-shaped, not plan-shaped.

Image-heavy work: Gemini AI Pro for published, predictable volume at $20; ChatGPT Pro if you want effectively uncapped generation and will pay for it.

And whichever you pick: check the live product page before subscribing. Every number in this post is a September 2026 snapshot of the fastest-changing prices in software — several of them changed in the four weeks before this was written.

Sources

• OpenAI Help Center: Model release notes
• OpenAI: Codex pricing and official model-specific usage ranges
• ChatGPT: Go plan page
• ChatGPT: Pricing page (Free/Go/Plus/Pro feature wording)
• BentoML: ChatGPT usage limits explained
• Carly: ChatGPT message limit — Free vs Plus vs Pro (2026)
• GPT Image: ChatGPT image limits by plan, 2026
• Claude Help Center: How do usage and length limits work?
• Claude Help Center: Usage limit best practices
• Carly: Claude usage limit — Free vs Pro vs Max (2026)
• Google: Gemini Apps limits & upgrades for Google AI subscribers
• Google: Changes to Gemini model access and limits
• 9to5Google: Gemini app now has compute-based usage limits as AI Ultra now starts at $100
• Android Police: Google is changing how Gemini usage limits work
• Carly: Gemini usage limits — Free, AI Pro & AI Ultra (2026)

Limit figures are the September 2026 snapshot labeled in each table (OpenAI's pricing-page wording was checked September 6, 2026); model changes through October 11, 2026 are noted in the update at the top. This post is re-checked monthly. Where vendors don't publish numbers, figures above are clearly labeled as user-reported approximations — your account's in-app meter or banner is always the source of truth. This article is informational, not purchasing advice.