ChatGPT can now watch your Gmail, log into websites for you, and cost less to run

Picture this: a pull request lands in your GitHub repo, and before you've even opened your laptop, ChatGPT has already summarized the changes and flagged anything that looks risky. Or an email from a client hits your inbox at 6am, and by the time you check your phone, there's already a draft reply waiting for your approval. That's not a hypothetical anymore — it's a feature that shipped this year, along with two other updates that quietly changed how useful ChatGPT can be for actual work.

OpenAI rolled out three updates recently that don't get the same headlines as a new model release, but they change the day-to-day experience of using ChatGPT more than most model upgrades do. One lets ChatGPT react to things happening in your other apps. One lets it log into password-protected websites without ever seeing your password. And one just made its most powerful model noticeably cheaper to run. Here's what each one actually does, and how to turn them on.

What are ChatGPT's new event-triggered tasks?

Up until now, ChatGPT's "Tasks" feature worked on a clock — you could schedule a prompt to run every morning at 8am, for example. The new update adds a second kind of trigger: instead of waiting for a specific time, ChatGPT can now wait for a specific event in Gmail, Slack, or GitHub and fire off a prompt the moment it happens.

Event-triggered (webhook-based) tasks run in Work and respond to supported Gmail, Slack, or GitHub activity, and they are available to eligible users on Plus, Pro, Business, Enterprise, and Edu plans. Free and Go accounts cannot create event-triggered tasks, and event-triggered tasks are not available in FedRAMP workspaces.

Here's what each integration can watch for:

• Gmail — tasks can respond to a new message or filter by sender or subject.
• Slack — tasks can respond to new channel messages, and you need to add @ChatGPT to each monitored Slack channel.
• GitHub — tasks can respond to supported pull request activity in an authorized github.com repository.

How do you set one up?

The setup lives inside ChatGPT itself, no separate dashboard required. According to OpenAI's help center article on scheduled tasks, the flow looks like this:

1. Go to Settings > Apps and connect a supported Gmail, Slack, or GitHub account.
2. Open Work and describe the event and the action ChatGPT should take.
3. Review Trigger, Condition, and Prompt, and complete any required authorization.
4. Go to Scheduled to view and manage tasks.

You don't need to write any code or configure a webhook manually — you just describe what you want in plain language, and ChatGPT translates that into a trigger, a condition, and a prompt for you to review before it goes live. If you want the more technical breakdown of how the trigger logic works, ChatGPT Learn's automations docs lays out the distinction between event-triggered and time-based tasks clearly.

What should you actually automate first?

Don't try to automate everything on day one. The people getting the most out of this feature are starting small and specific rather than broad. A few starting points worth trying:

• Gmail — trigger a summary whenever an email from a specific client or vendor lands, so you get the gist without opening your inbox.
• Slack — have ChatGPT draft a follow-up checklist whenever a teammate posts in a project channel that a milestone is done.
• GitHub — kick off a summary of open PR comments before your team's morning standup.

One practical note: this is genuinely new behavior for ChatGPT, not just a bigger version of the old scheduled tasks feature. Until August 25, autonomous scheduled tasks required a paid subscription, and the free-tier addition represents the first time OpenAI has brought any form of autonomous, unattended task execution to its largest user tier. The event-triggered version is still paid-plan only, but it's the same underlying shift toward ChatGPT acting on its own instead of waiting to be asked.

How does ChatGPT log into websites without seeing your password?

The second update solves a problem that's been holding back AI agents for a while: how do you let an assistant complete tasks on password-protected sites — booking a DMV appointment, checking a reimbursement, setting up utilities for a new apartment — without handing it your actual login credentials?

ChatGPT Work can now sign in to websites on web and mobile using its computer and browser, without ChatGPT seeing the user's username or password, which allows it to complete browser-based tasks that require authenticated access.

Here's how the handoff actually works in practice. You ask it to do something on a site that needs a login, and if that site allows authentication, ChatGPT surfaces the login screen so you can type your credentials or a security code — once you're in, the agent keeps going and the session may stay signed in for future tasks. Screenshots are disabled during takeover, so OpenAI never sees your password.

If you use a password manager, the experience gets even smoother. OpenAI says the browser supports password managers for that step, and that the model can't see your username or password; they aren't stored and aren't used for training.

What can you actually use this for?

OpenAI's own examples give a good sense of the intended use cases: setting up utilities for a new apartment, booking a DMV or passport appointment, and checking a reimbursement. Basically, the tedious admin tasks that require an account login but don't require much judgment — the stuff you'd normally do yourself purely because it needs a password, not because it needs a human brain.

What's the catch you should know about?

This is genuinely useful, but it's worth understanding what the safeguard does and doesn't cover. Keeping the password out of the model's view is a real security improvement, but it doesn't eliminate risk entirely. An agent that can log in and stay logged in is a useful assistant and a standing credential — the safeguards OpenAI lists cover the password, but they don't cover the session the password unlocks. In plain terms: your password stays hidden, but the logged-in session itself still has real access to your account, so it's smart to review what the agent is doing before it takes any action that matters — payments, account changes, or anything that can't easily be undone. Consequential actions still require review, and ChatGPT is designed to ask before actions that create a financial, legal, account or other real-world commitment.

If you're rolling this out for a team rather than personal use, it's worth checking Settings > Cloud browser to manage sessions per site, and Settings > Security for additional lockdown controls before letting the agent loose on anything sensitive.

How much cheaper did GPT-5.6 Sol get?

The third update is less about new capability and more about making existing capability cheaper to use at scale. OpenAI cut the price of its flagship model, GPT-5.6 Sol, across the API, Codex credits, and ChatGPT Work. OpenAI is dropping API and credit pricing of GPT-5.6 Sol by over 20% for the next 3 months, starting with the same Sol intelligence at the lower rate.

The actual numbers, if you're comparing token costs: GPT-5.6 Sol now costs $4 per 1 million input tokens and $20 per 1 million output tokens for standard short-context use, down from $5 and $30, with the promotional rate guaranteed at least through November 21. That's roughly a 20% cut on input and a 33% cut on output — a meaningful difference if you're running Sol through heavy Codex workloads or long agentic sessions in ChatGPT Work.

Worth noting: pricing for Pro, Plus and Business subscriptions is unchanged, and the lower metered rates do not add capacity to those plans. So if you're on a flat-rate consumer plan, this update mostly matters if you're also using the API or Codex credits directly. If you're a developer or team paying by usage, though, it's a straightforward win — the same model, doing the same work, for noticeably less.

You can check current pricing directly on OpenAI's platform pricing page, and the original announcement is up on the OpenAI Developer Community forum if you want the full details straight from the source.

Which of these updates should you try first?

If you're drowning in Slack notifications or email triage, start with event-triggered tasks — it's the update that saves the most time for the least setup effort. If you spend real hours a week on account-gated admin (utilities, appointments, reimbursements), the secure sign-in feature is worth testing on one low-stakes task before trusting it with anything bigger. And if you're a developer running heavy workloads through the API or Codex, the price cut is free money — no setup required, just lower bills.

None of these three updates are flashy in the way a new model launch is. But together, they point at something bigger: ChatGPT is quietly moving from a tool you talk to, into a system that watches, acts, and logs in on your behalf. The question worth sitting with isn't which update is coolest — it's how much of your actual workflow you're comfortable handing over next.