ChatGPT Plus limits: messages, images, files and Codex caps

ChatGPT Plus caps your usage over two windows at once: a rolling five-hour message window plus a separate weekly allowance on Work and Codex. On OpenAI's published estimates, the $20-a-month Plus plan buys 5 to 45 local GPT-6 Astra messages per five-hour window, 15 to 150 on GPT-6 Sol, or 350 to 3,000 on GPT-6 Luna, and images draw the same pool down 3 to 5 times faster than plain messages 1. Pro multiplies every allowance: the $100 tier carries 5x Plus usage, the $200 tier 20x 1. The counts are estimates rather than fixed entitlements, and OpenAI's usage dashboard is where your real numbers live 1.

The five-hour window: one shared pool for ChatGPT Work and Codex

ChatGPT Work and Codex do not have separate caps. OpenAI's pricing documentation states it flatly: Work usage inside ChatGPT uses the same pricing, credits, and usage limits as Codex 1. Burn the pool in the Codex CLI and your Work sessions in the desktop app feel it, and the reverse. Local messages and cloud chats draw on the same allowance, and cloud chats run on GPT-5.6 Sol while tending to consume more of the allowance per task 1.

The five-hour estimates OpenAI publishes, per model, on Plus 1:

• GPT-6 Astra: 5 to 45 local messages
• GPT-6 Sol: 15 to 150
• GPT-6 Luna: 350 to 3,000
• GPT-5.6 Sol: 10 to 100 (also the model behind cloud chats)
• GPT-5.6 Terra: 25 to 200
• GPT-5.6 Luna: 250 to 2,000

The spread inside each range is the whole story: model choice, context size, reasoning effort, tool use, and caching all change how much one message costs, so tasks that look similar consume different amounts 1. One seam trips Plus subscribers in particular. GPT-6 Astra reaches Plus only through Work and Codex experiences; the Astra-powered option in ordinary Chat is named GPT-6 Pro and sits with the Pro plans 2, a split Plus users have pushed back on in OpenAI's own community forum 3. If your question is which models appear in the Chat picker rather than how far the Work and Codex window stretches, that is a model-availability question, and availability varies by workspace the way OpenAI's Business models and limits article describes 4.

The weekly window is the cap the pricing docs do not print

The five-hour table is only half the system. Weekly limits may also apply, in OpenAI's own wording, and the weekly allowance shows up only in your usage dashboard alongside its reset date; the pricing documentation prints no weekly message counts for any plan 1. The reset mechanics are documented, though. A paid instant reset, where offered on personal plans, brings the normal weekly allowance forward immediately, and the next weekly reset lands seven days after your first subsequent Work or Codex request 2. Banked resets, granted through promotions and referrals, stay usable for 30 days 1. The complete reset windows and credit rules per plan are tabulated in our credits-reset guide 5.

Images, voice, files, and deep research: what shares the pool

Image generation carries no separate daily counter in the pricing documentation; it draws the same general limits as messages and burns them 3 to 5 times faster on average, with quality and size setting the exact rate. Images are not available on the Free plan, and once included limits run out, image generation draws from credits 1. Desktop voice bills the shared budget at $0.05 a minute, or 1.25 credits a minute on credit-based billing, the rate that also applies when Plus and Pro users spend additional credits 1.

File uploads keep their own rolling cap, separate from the message pool. OpenAI's help center does not expose that article to automated reads, and the most recent third-party documentation of the exact counts, from August 2025, puts Plus at 80 files per rolling three hours and Free at roughly 3 a day 6; a current tracker lists Free at 5 files per project 7. Treat the dashboard as authoritative. Deep research queries carry per-month counts: 25 on Plus, 250 on Pro, and 5 on Free through a lightweight variant of the feature, the per-tier figures from OpenAI's April 2025 increase, the last numbers the announcement page prints 8.

Credits are the overflow valve across all of this: after included limits are reached, credits extend usage for supported features 9, GPT-5.6 usage averages 5 to 30 credits per message 1, and credit purchase prices depend on your plan or agreement rather than a single posted rate 1.

Where the five-hour window actually breaks

The window is a compute budget wearing a clock face. A Plus user working a serious repository documented the allowance draining in about an hour, with project and context loading consuming budget before useful work starts, and asked OpenAI to stop presenting a token budget as hours 10. OpenAI's own behavior at the wall softens the landing: an in-progress turn finishes, subject to fair-use limits, and Plus and Pro users can buy credits to keep working without changing plans 1.

Two levers change how far the window stretches. Speed is one: Fast mode runs GPT-6 Astra, Sol, and Luna at 2.5x the Standard credit rate 11, which by simple division turns the 5-to-45 Astra range into roughly 2 to 18 effective messages. Model choice is the bigger one: GPT-6 Luna's Plus range is about 70 times GPT-6 Astra's at both ends of the estimate, the largest single lever inside the same $20 1. OpenAI's stretch advice runs the same direction: trim prompts, limit source material, cut MCP servers, shrink AGENTS.md context files 1.

What each plan pays per unit of allowance

The ladder, as a delta 7:

• Every plan, Free included, carries the Chat surface; Free gets unlimited everyday text chats on GPT-5.6 Luna 12, ad-supported in the United States 13.
• Go, $8 a month, adds roughly 10x Free capacity and wider upload and image limits, still ad-supported 7.
• Plus, $20 a month, removes ads and adds Work and Codex with Astra access, deep research, and credit purchases 7 1.
• Pro at $100 multiplies usage 5x and adds a larger context window, the top reasoning mode, and GPT-6 Pro in ordinary Chat 7 1.
• Pro at $200 multiplies usage 20x, with the highest reasoning allowance and maximum tool throughput 7.
• Business uses the Pro 5x usage estimates in its $100 configuration, and the Standard Business estimates in the same table equal the Plus ranges; every seat adds workspace admin controls 1.
• Enterprise and Edu plans with flexible pricing carry no fixed rate limits; usage scales with credits, and without flexible pricing their per-seat limits match Plus for most features 1.

Divide each price by its multiple and the ladder tells you something the pricing pages do not print. Plus is $20 for 1x. Pro at $100 is $20 per Plus-equivalent unit, exactly the same rate; what that tier sells is the context window and reasoning mode, not cheaper capacity. Pro at $200 is $10 per unit, half of Plus, and it is the only rung where the per-unit price falls 71. Heavy Astra and Codex users are the readers this arithmetic moves most.

Codex run with an API key skips the windows entirely and bills at API token rates; GPT-6 Astra lists at $10 per million input tokens and $50 per million output on the API, prices that do not convert into subscription allowance 2 1. Our OpenAI Codex pricing guide covers that meter, and every credit, in detail 14.

Who the Plus limits fit, and when to move

Plus fits work that stays inside roughly 40 flagship messages per five-hour window with room in the weekly allowance. Hit the five-hour cap occasionally and the fixes are cheap: drop to Sol or Luna for routine turns, keep Fast mode off, or buy credits for the burst. Hit the weekly cap week after week and the ladder is blunt: Pro at $100 if you also want the context and reasoning upgrades at the same per-unit price, Pro at $200 if the need is purely volume, because that is where capacity gets cheaper. If the move you're weighing is across vendors rather than up OpenAI's ladder, our Google AI Pro vs ChatGPT Plus comparison prices both sides 15. Image-heavy and voice workloads should budget against the 3-to-5x burn and the per-minute voice rate before assuming a month of headroom. Free and Go readers whose needs are text-shaped already have unlimited everyday chats on Luna; the caps that bite them are premium models and agentic features 12. To test which rung pays for itself against your actual usage pattern, run the break-even with your own numbers in ProvenBrief's AI subscription calculator 16.

Verified September 2026 against OpenAI's Codex pricing and usage documentation, its ChatGPT Learn docs, and current independent plan trackers. The vendor's help-center articles and consumer pricing pages refuse automated reads; those figures were checked through search-index records of the vendor's own pages and are marked with their dates. This page is maintained: when OpenAI changes a cap, the figure above and a dated change note update here.