ChatGPT's Model Picker Just Got a Rethink: Here's What the New Effort Tiers Mean for You

If you’ve been a ChatGPT Plus or Pro subscriber for a while, you’ll know the model picker has gone through a few iterations. First it was a list of versioned model names. Then OpenAI introduced Auto, Fast, and Thinking modes. Now there’s another change rolling out, and this one is arguably the most straightforward yet.

OpenAI is replacing named-model selection with a set of effort-based tiers. Instead of picking a specific model, you choose how hard you want ChatGPT to think.

What the New Tiers Look Like

The updated picker offers up to six options, depending on your subscription level:

• Instant — fast responses for simple tasks. Think quick lookups, short rewrites, or anything where speed matters more than depth.
• Medium — a step up in reasoning, suitable for most everyday tasks that need a bit more than surface-level thinking.
• High — deeper reasoning for more complex questions or multi-step problems.
• Extra High — pushes the model harder still, for demanding analytical work.
• Pro Standard and Pro Extended — reserved for Pro subscribers, these represent the heaviest computational lift available and deliver the highest reasoning capability at the cost of longer response times.

If you were using the previous naming system, here’s the direct translation: Thinking Standard is now Medium, Thinking Extended is now High, and Thinking Heavy is now Extra High. One option that doesn’t carry over is Thinking Light, which OpenAI has removed. According to OpenAI, fewer than 1% of paid users were using it, so the decision to cut it is fairly unsurprising.

Where to Find It

On web, the picker sits directly in the message composer. On iOS and Android, you’ll find it at the top of the conversation. The rollout is global for Plus and Pro subscribers across all three platforms.

The Auto-Escalation Setting

One detail worth knowing about: when you select Instant, ChatGPT can automatically step up to Medium if it determines your query needs more reasoning. This behaviour is controlled by a toggle in Settings > General, and you can turn it off if you prefer to stay in manual control of which tier is active.

This is a sensible design choice. Instant is genuinely useful for quick tasks, but you probably don’t want it fumbling through something complex just because you forgot to switch tiers. The auto-escalation covers that without requiring you to think about it every time.

If you’re the type who wants precise control over compute and response time, turning it off gives you that.

Why OpenAI Made This Change

The shift from model names to effort tiers reflects a broader simplification that OpenAI has been working toward across its product line. As the GPT-5 family expanded, picking between specific model versions became increasingly opaque. Most users don’t have a strong reason to care whether they’re on GPT-5.5 versus a reasoning variant of it. What they care about is whether the response will be fast, or thorough, or somewhere in between.

Framing the choice as effort rather than model identity maps more directly to how people actually think about the task in front of them.

OpenAI has also said it factored in community feedback during this redesign, specifically to ensure that the higher-end options like Extra High, Pro Standard, and Pro Extended are easy to find and clearly labelled, and that the transition away from the old naming is communicated clearly. The ChatGPT release notes page on the Help Center has the full details.

What This Means for You

For most Plus users, the practical day-to-day change is small. You’ll see different labels, but the underlying capability is the same. Instant handles the quick stuff, Medium covers most general use, and High is there when you’re working through something that genuinely requires more reasoning.

The auto-escalation toggle is the one setting worth checking early. If you use Instant as a default and want ChatGPT to quietly upgrade when needed, leave it on. If you’re trying to manage response times or have a specific reason to stay on lighter compute, turn it off.

For Pro subscribers, the new layout makes Pro Standard and Pro Extended more prominent than they were previously. If you’ve been on Pro and haven’t regularly used the heavy reasoning modes because they were buried, that changes with this update.

Free users don’t get access to the new picker. The effort tiers are a paid feature across Plus and Pro.

The removal of Thinking Light is the only genuine subtraction here. If you were relying on it, the closest equivalent is now Instant, though they’re not identical. Given the usage numbers OpenAI cited, this will affect very few people.

Overall, this is a clean simplification. The model picker has been one of those interface elements that accumulated complexity over time, and this redesign resets it around something more intuitive. Choosing how hard to think rather than which model to use is a more honest representation of what you’re actually deciding.

---

Updates to this story

12 July 2026: OpenAI temporarily removes the 5-hour usage cap and resets quotas for Plus, Pro, and Business after GPT-5.6 Sol demand surge

GPT-5.6 Sol doubled ChatGPT traffic within 48 hours of launch, and the effort-tier system described in this article felt the pressure immediately. Higher-effort tiers consume more tokens per session, and because Codex and ChatGPT Work share the same usage pool, many users hit their limits rapidly. On July 12, OpenAI product lead Tibo Sottiaux announced three emergency changes in response.

The rolling five-hour usage cap is temporarily lifted for Plus, Pro, and Business plans, with no confirmed return date. A one-time quota reset was applied to all affected accounts, and OpenAI is expanding its “banked reset” feature, previously desktop-only, to web and mobile for around 500,000 ChatGPT Work and Codex users.

On the efficiency side, inference optimizations should deliver roughly 10% more effective usage from GPT-5.6 Sol. A separate issue, where Sol’s context window had been set to 372k tokens instead of the intended 272k, was also corrected, meaning usage should drain noticeably slower going forward.

Weekly caps still apply, so the service is not unlimited. OpenAI has also reverted some internal reasoning-effort adjustments that were contributing to higher-than-expected consumption. BleepingComputer has a solid summary of the full rollout.

17 June 2026: ChatGPT Scheduled Tasks overhaul: dedicated Scheduled page, flexible time windows, and monitoring tasks that ping you only when something changes

OpenAI has shipped a significant upgrade to ChatGPT’s scheduled tasks feature, available now to Plus, Pro, Business, and Enterprise users on web and mobile (not the desktop or Codex apps).

The headline addition is a dedicated Scheduled page in the sidebar, where all active tasks are listed alongside their next run times. From there you can pause, edit, resume, or delete any task without hunting through old conversations. Scheduling itself is now more flexible: you can pin a task to a specific time or give ChatGPT a broader window, such as morning, afternoon, or evening.

The more consequential new capability is monitoring tasks. Rather than running on a fixed schedule regardless of what has changed, a monitoring task checks connected apps or the web periodically and notifies you only when something worth reporting has actually happened. Previous runs are remembered, and the task can stop automatically once an end condition is met.

Active task limits vary by plan: up to 5 for Plus, up to 15 for Pro, Business, and Enterprise. Tasks cannot run more than once per hour. Separately, OpenAI is retiring Pulse, with Pro users retaining access for 14 days before daily briefings move into the scheduled tasks system.

This extends the effort-tiers story: where those tiers changed how ChatGPT reasons on demand, scheduled tasks change when it acts, nudging ChatGPT toward a persistent background layer rather than a tool you summon.