ChatGPT removes automatic Thinking for Plus and Pro: what to choose now

OpenAI is retiring automatic switching from Instant to Thinking for ChatGPT Plus and Pro users worldwide, and removing the Higher intelligence setting from the web version of those plans. The change, announced in the company's [September 14 release notes](https://help.openai.com/en/articles/6825453‑chatgpt‑release‑notes), leaves manual reasoning options available. A complex request submitted with Instant selected will no longer receive a higher thinking level simply because ChatGPT decides it needs one. Automatic switching for safety remains possible. The practical consequence is a change in how subscribers prepare a demanding request. Choosing a reasoning level now matters before sending it; adding an instruction to think more deeply does not change the selected level. Business and Enterprise users need a different explanation, because their automatic‑routing controls have not been covered by the same personal‑plan retirement. OpenAI's [current model guidance](https://help.openai.com/en/articles/20001354) and its separate workspace documentation describe three different situations behind what can look like the same missing setting.

For personal Plus and Pro accounts, OpenAI explicitly says that asking ChatGPT to think harder or think deeply does not automatically switch the conversation to another thinking level. The supported route is to choose an available option in the model picker. On eligible paid plans, GPT‑5.6 Sol powers Instant, Medium, High and Extra High, with Medium providing standard reasoning and High extended reasoning. The Pro option uses separate Pro models. These distinctions are set out in the [model and thinking‑level documentation](https://help.openai.com/en/articles/20001354). Consider a subscriber comparing several proposals against a set of requirements. The request may contain substantial context and an instruction to examine trade‑offs carefully, but its wording is not a substitute for selecting a reasoning level. Choose the level before sending the comparison, then supply the proposals, constraints and desired output. Once the response arrives, check its claims against the proposals and requirements, including whether it has dropped a constraint or attributed a feature to the wrong proposal. There is also no reason to treat the missing Higher intelligence web setting, by itself, as evidence that a Plus or Pro subscription has lost manual Thinking access. OpenAI's [availability table](https://help.openai.com/en/articles/20001354) gives Plus users Medium and High, while Pro users also have Extra High and Pro. That makes the first check specific: which plan is active, and which reasoning options does its picker show? Searching for the retired personal‑plan switch will not restore the former automatic behavior.

>>ChatGPT Business retains an automatic‑reasoning control under Settings General called Higher intelligence. Its [Business documentation](https://help.openai.com/articles/12003714) says ChatGPT can add reasoning while Instant remains selected, and that this can consume the reasoning allowance. A person using both a personal account and an employer's workspace can therefore encounter two valid interfaces: the switch is absent on personal Plus or Pro web, while Business still documents it. The September 14 announcement should not be applied to every paid workspace. That distinction matters when a Business user is trying to understand usage. Seeing Instant selected is not sufficient evidence that a request used only the base chat allowance. Check the workspace's Higher intelligence setting together with its current allowance information. Business also distinguishes Standard and Premium seats from the Pro model option: Premium provides more included Work and Codex usage, while individual Chat model allowances are separate. A seat label alone does not explain either the reasoning used for a particular request or the remaining allowance. | Account or workspace | Current control | What to check first | | --- | --- | --- | | Personal Plus | Manual Medium or High; Higher intelligence removed on web | Select the available reasoning level before sending a complex task | | Personal Pro | Manual Medium, High, Extra High or Pro; Higher intelligence removed on web | Selected level and the relevant model's remaining allowance | | Business | Manual options plus Higher intelligence under Settings General | Automatic reasoning setting, workspace access and allowance | | Enterprise or Edu | Workspace‑controlled options and possible admin routing controls | Enabled models, assigned roles and rollout status | | Free or Go | Think uses GPT‑5.6 Luna | Whether the expected option belongs to the user's plan | The table combines OpenAI's [personal‑plan model guide](https://help.openai.com/en/articles/20001354), [Business rules](https://help.openai.com/articles/12003714) and [Enterprise and Edu controls](https://help.openai.com/articles/11165333). It separates the action a reader can take from the options an organization may allow. Free and Go users, for example, can use Think for harder questions, but do not have GPT‑5.6 Sol access; the absence of a paid‑plan Sol option does not mean their Luna‑based Think feature has been withdrawn.

>Enterprise and Edu administrators may have automatic‑routing controls under Workspace settings Models, with labels and availability varying by workspace and rollout. Members can still choose reasoning options enabled for them. OpenAI also says some Enterprise customers may continue to see an older picker while the new controls arrive. Those qualifications in the [managed‑workspace documentation](https://help.openai.com/articles/11165333) make a screenshot from another account an unreliable checklist for whether an employee's own configuration is correct. A default model and permission to use it are separate controls. Choosing a starting model does not grant access that a member lacks. For GPT‑6 Astra, OpenAI says access is off by default at launch in eligible Enterprise and Edu workspaces, and previous Early Model Access settings do not automatically carry over. Assigned roles matter too: a role allowing access can keep it enabled even if another role disables it. These are access rules, separate from whether Instant automatically sends a request to reasoning. An administrator investigating a discrepancy can use Model Test in the Admin console: open the ChatGPT workspace, select Models, then Test, and search for the member. The result shows effective access and the settings contributing to it, including roles, permissions and preferences. OpenAI's [Model Test instructions](https://help.openai.com/articles/11165333) say the check reads saved settings; it does not enable a model, increase a limit or override eligibility. That provides a concrete way to diagnose a missing option before repeatedly changing defaults or asking the employee to rewrite the prompt.

Manual Thinking and the Pro model options have usage rules that the new interface does not erase. OpenAI says that choosing Medium, High or Extra High uses GPT‑5.6 Sol, and that reaching a Thinking limit may cause ChatGPT to continue with another available model. Its [current limits guide](https://help.openai.com/en/articles/20001354) also separates Pro‑model allowances from the Pro subscription itself. Reaching a model allowance does not change the subscription, and ChatGPT shows a reset time when that information is available. For $200 Pro, GPT‑6 Pro has a 200‑message weekly allowance. GPT‑5.6 Sol Pro has a separate 170‑message daily allowance, and both models together have a 200‑message daily ceiling. Reaching the GPT‑6 Pro weekly limit switches the conversation to Medium; using Sol Pro instead requires selecting GPT‑5.6 Sol and then Pro, with room remaining in both relevant daily allowances. The $100 Pro plan works differently: its two Pro models share 50 messages a week, so moving between them cannot create another allowance. These are the [published Chat limits](https://help.openai.com/en/articles/20001354), separate from Work and Codex. Business has another distinction: Standard includes 15 Pro messages a month, while Premium includes 50 a week, shared across GPT‑6 Pro and GPT‑5.6 Sol Pro. Switching models does not reset that included allocation, according to the [Business limits page](https://help.openai.com/articles/12003714). To investigate, establish the workspace, confirm access to the model, identify the selected level, then inspect the applicable allowance. Each step explains a different reason that an expected option or behavior might be unavailable.

ChatGPT's controls have changed several times this year. An [April 30 TechRadar guide](https://www.techradar.com/ai‑platforms‑assistants/chatgpt/chatgpt‑just‑made‑it‑easier‑to‑pick‑the‑right‑model‑just‑like‑gemini‑does‑heres‑when‑to‑use‑instant‑thinking‑or‑pro) described an automatic‑selection switch and older model labels. Its [June 19 mobile guide](https://www.techradar.com/ai‑platforms‑assistants/chatgpt/i‑found‑a‑hidden‑chatgpt‑setting‑that‑changes‑how‑hard‑the‑ai‑thinks‑and‑the‑difference‑surprised‑me) covered a gesture‑based route to choosing intelligence levels. Those were accounts of earlier interfaces. Neither should be read as evidence that the September 14 personal‑plan automatic switch still exists, or that every current workspace must show the same controls. For a personal Plus or Pro user, the immediate adjustment is to choose an available thinking level when preparing a task that warrants more reasoning. For a Business user, the additional check is whether Higher intelligence is changing behavior while Instant stays selected. For an Enterprise or Edu member, ask the administrator to inspect effective model access. If the problem is instead an exhausted allowance, waiting for its reset or choosing another available model follows the documented rules; it is separate from the subscription decisions covered in our [guide to $200 Pro signups and billing](https://opentools.ai/news/chatgpt‑pro‑200‑signup‑pause‑existing‑subscriptions‑billing). The missing switch, an absent model and a usage cap can produce similar frustration, but they call for different actions.