The Complete Guide to Tracking Your ChatGPT Usage in 2026

There are four ways to track ChatGPT usage in 2026: the OpenAI API dashboard at platform.openai.com (API users only), browser extensions that overlay a counter on chatgpt.com, Mac menu-bar apps like tokn.watch that poll OpenAI's internal rate-limit endpoint, and a manual DevTools probe of that same endpoint for one-off checks. The single most important fact: ChatGPT does not have a built-in usage page. Unlike Claude, you cannot just open Settings and read a percentage - you have to install something or dig. Here is what each option shows, what it does not, and which one to pick.

Update, 2026-09-18: OpenAI released GPT-6 Astra on 2026-09-03. Regular text chat is still uncapped and still runs on GPT-5.6 Sol for Plus (Luna on Free and Go). Astra itself is limited: Plus gets it only in ChatGPT Work and Codex, at roughly 5-45 messages per rolling 5-hour window. Pro gets it in Chat as GPT-6 Pro with a weekly allowance (50 messages on the $100 tier, shared with GPT-5.6 Sol Pro; 200 on the $200 tier), plus roughly 25-225 ($100) or 100-900 ($200) Work messages per 5 hours. Free and Go have no Astra access. OpenAI also paused new sign-ups and upgrades to the $200 Pro tier on 2026-09-11, with no end date announced; existing $200 subscribers and the $100 tier are unaffected. Those Astra allowances are the new thing worth tracking.

The four ways to track ChatGPT usage, side by side

The honest starting point is that OpenAI does not want you to look. Claude has a Settings then Usage page that surfaces every percentage; ChatGPT has nothing equivalent on chatgpt.com. Text chats have been uncapped since 2026-08-06, but for the allowances that remain (images, files, Deep Research, Agent mode, Codex sessions) you still find out you have hit one by hitting it: a banner appears, with a reset time. There is no preemptive bar, no countdown, no notification, no "you have used 80% of this window" warning. This is true on Free, Plus, Pro, and Team. It is the reason every tool on this list exists.

Here is the shape of each option, the question it answers, and the trade-off it makes.

Three of these are something you actively check. The third is something you glance at while you work. That difference matters more than any feature comparison. If you only realise you should have looked at usage after you have already hit the wall, you do not need a better dashboard; you need an ambient signal.

Below is the detail for each option, with the gotchas the marketing copy does not mention.

Method 1: The OpenAI API dashboard at platform.openai.com

The first thing most people find when they search "how to check ChatGPT usage" is platform.openai.com. This is the right answer for the wrong question. The dashboard is for OpenAI's API product - it shows API request volume, dollar spend by model, and rate-limit headroom for accounts that hold an API key. It is unrelated to your ChatGPT Plus or Pro subscription.

If you only use chatgpt.com (the consumer app), the dashboard will show zero usage even on a heavy week. The reason is structural: ChatGPT Plus / Pro and the OpenAI API are separately billed, separately metered, and separately surfaced. OpenAI's own help-center article on ChatGPT Plus confirms this: the subscription gives you access to GPT-5.6 Sol, Sora, and so on through chatgpt.com - none of which routes through the API metering at platform.openai.com.

Two things the dashboard does show that are useful even if you only use chatgpt.com:

• Whether you accidentally have an API key in active use. If you ever signed up for the API as a developer and forgot, the dashboard will show running costs. Some users discover here that an old script or a colleague is silently billing their card.
• Rate-limit tiers, which OpenAI documents publicly. Useful background if you ever build a tool that calls the API.

For consumer ChatGPT usage tracking, skip this one. Look at it once to confirm you have no surprise API spend, then move on.

Method 2: Browser extensions that count messages in the page

The second class of tool runs as a browser extension and overlays a counter directly on chatgpt.com. Two are popular enough to dominate the search results on the Chrome Web Store in 2026:

• ChatGPT Usage Limit Tracker - injects a small floating widget showing your current message count for the active model, a reset timer, and a daily / weekly breakdown.
• ChatGPT Token Counter - shows estimated token usage rather than message count; more useful for users on the API but also runs on chatgpt.com.

The good news: extensions are zero-friction to install, free, and they show numbers immediately on the page you already use. The not-so-good news is structural.

Extensions count messages they have observed since the tab was opened. They do not poll OpenAI's internal endpoint. If you close the tab, the counter resets. If you switch browsers (Chrome at the office, Safari at home), each browser has its own count, neither aware of the other. If you also use ChatGPT in the desktop app or on mobile, the extension misses every message sent there. The result is that the on-screen "remaining" number drifts further from reality the more places you use ChatGPT.

They also fight with chatgpt.com's own DOM. Every time OpenAI ships a UI rewrite, the extensions need an update to keep finding the right hooks. Long stretches of "extension broken after the Oct redesign" reports are normal in this category - it is the price of scraping a UI that was never designed to be scraped.

Use a browser extension when you only use ChatGPT in one browser, on one device, for one account, and you do not mind that the counter resets every time you close the tab. For anyone else, the architecture creates more confusion than it removes.

Method 3: Menu-bar apps that poll the internal rate-limit endpoint

The third class of tool runs in your macOS menu bar, polls OpenAI's internal rate-limit endpoint every few minutes using your existing chatgpt.com browser session, and shows a live remaining percentage and reset countdown - no install on the website, no dependency on which browser you have open, no reset when you close a tab.

This is the only category that delivers an ambient signal for ChatGPT. You can glance up and see "62% W" without breaking your flow, and you can do it while ChatGPT itself is not even open.

In 2026 there is exactly one free Mac menu-bar app that covers ChatGPT:

• tokn.watch - tracks both ChatGPT and Claude in one menu-bar item. Free. Open source. Reads your existing browser session for both providers. Runs on macOS 13 and up, Apple Silicon and Intel. No account, no API key, no servers in the middle.

The Claude-only equivalent - ClaudeUsageBar - does not support ChatGPT, which is one of the structural reasons tokn.watch exists at all. The "Claude plus ChatGPT in one menu-bar item" niche has no other free entrant. Pricier paid alternatives exist on the Mac App Store that wrap a chatgpt.com web-view with notifications around the OpenAI banner, but they do not poll the rate-limit endpoint directly and the numbers are not live.

The trade-off menu-bar apps make: they need to authenticate with OpenAI's backend on your behalf. tokn.watch handles this by reading your existing chatgpt.com session cookie - the same one your browser already uses - so it does not ask for an API key, and it does not move your credentials anywhere off-device. The cookie is read once on first run and stored in macOS UserDefaults; the app talks to OpenAI directly and does not call any tokn.watch server.

Method 4: The DevTools manual probe

The last option is the one nobody recommends but everybody discovers eventually: just read the JSON yourself. ChatGPT's web app makes a call to its internal rate-limit endpoint on most page loads, and that call returns a JSON object containing your remaining message count for the current window and the reset timestamps.

The path is:

1. Sign in to chatgpt.com in Chrome, Edge, Brave, or any Chromium browser.
2. Open DevTools (Cmd-Opt-I on Mac) and switch to the Network tab.
3. Send any message in ChatGPT.
4. Filter the Network panel for limit or rate. Look for a request to /backend-api/conversation_limit or /backend-api/sentinel/chat-requirements.
5. Click the request, open the Response tab, and read the JSON. It includes your remaining quota and the reset time for each model class (GPT-5, GPT-5 Thinking, GPT-5 Pro on Pro accounts).

This is the same endpoint tokn.watch polls automatically. Reading it once by hand is a useful exercise if you want to understand exactly what is being tracked - and it is occasionally faster than installing a tool if you only need to check once a year. As a daily workflow it is unbearable. The same numbers are a glance away in a menu-bar app.

It is also worth knowing that the exact endpoint name and the JSON shape change every few months as OpenAI iterates. Tools like tokn.watch absorb those changes in updates; a hand-rolled DevTools probe stops working until you find the new endpoint name yourself.

How to choose between them

The right tool depends on three things: whether you use the OpenAI API, whether you use ChatGPT in more than one browser or device, and whether you need ambient awareness or just on-demand verification.

Decision shortcuts that hold for most people in 2026:

• You only use chatgpt.com, in one browser, on one machine. A browser extension is fine. Pick ChatGPT Usage Limit Tracker. Accept that the counter is approximate.
• You use ChatGPT across browsers, devices, and the desktop app. Use tokn.watch. The menu-bar app polls the server, so it does not care which surface you sent your messages from.
• You also use Claude. Use tokn.watch. It is the only menu-bar app that tracks both providers in one slot; running two separate tools is friction that erodes the value of the ambient signal.
• You are burning through the weekly allowance and want a warning before it bites. Use tokn.watch. The ambient remaining percentage is the warning - you will see it falling and pace yourself.
• You actually use the OpenAI API as a developer. Use platform.openai.com for the API side. For the consumer ChatGPT side on top of that, layer on tokn.watch or an extension - the dashboard tracks neither.

It is also fine to combine. Many heavy users keep tokn.watch running for the ambient signal and open platform.openai.com once a month to reconcile API spend against the credit-card statement. The tools are not mutually exclusive; they measure different things.

Three things none of these trackers tell you

Even with all four tools running, there are gotchas in ChatGPT's usage model that no current tracker surfaces well. Three of them are worth knowing about:

1. The remaining caps are silently feature-specific. Since 2026-08-06 text chat is uncapped on every plan, but ChatGPT still runs separate windows for Sora, image generation, Deep Research, Agent mode, and (on Plus, since 2026-08-26) Codex and ChatGPT Work sessions. You can be fine on one and at 4% remaining on another. Most trackers show a single number labelled "weekly" - it is whichever window the tracker decided to surface. tokn.watch surfaces the windows your account actually reports.

2. The Free tier used to downgrade you mid-conversation - not any more. Until August 2026, when a Free user ran out of flagship quota, ChatGPT silently switched to a lighter model (GPT-4o mini) without flagging the switch, and the quality dived. Since 2026-08-06, Free text chats are uncapped on GPT-5.6 Luna, so this failure mode is gone for text - but image and file allowances on Free still run out silently.

3. ChatGPT Pro's "unlimited" is unlimited within fair-use limits that OpenAI does not publish. Pro at $200/month is marketed as unlimited GPT-5.6 Sol Pro access. (GPT-6 Pro, added 2026-09-03, is the exception: it carries a stated weekly allowance of 200 messages on the $200 tier and 50 on the $100 tier.) In practice there is a fair-use clamp that triggers on accounts that look automated - sustained 24/7 calling, parallel scripted requests, content-farm patterns. The clamp is not surfaced in any UI and rate-limit endpoints do not return a meaningful number for Pro accounts. If you are a normal interactive user, this never matters; if you are running an agent that calls ChatGPT in a loop, you will eventually find the wall the hard way.

For the full mechanics of the underlying caps and what trips each one, see how ChatGPT's usage limits really work in 2026.

What the numbers look like compared to Claude

If you came to this article from a Claude background, the headline difference is that Claude shows you the percentages and ChatGPT does not. Claude's Settings then Usage page is a clean dashboard with current session and weekly numbers, plus a Fable weekly meter on Max plans. ChatGPT has no equivalent surface, on any subscription tier, on any platform.

That structural gap is why the ChatGPT tracker ecosystem is sparser and weirder than the Claude one. Claude has four official-ish ways to check (Settings, /usage, ccusage, menu-bar apps). ChatGPT has one tool category that polls (menu-bar apps), one that scrapes (browser extensions), one that is mis-targeted (the API dashboard), and one that is manual (DevTools). It is the same set of trade-offs the Claude side has, but with the official option missing.

For the Claude side of the same story, see the complete guide to tracking your Claude usage.

Frequently asked questions