Instant
Use for clear, familiar tasks where speed matters: emails, summaries, ideation and quick edits.
Models & Reasoning
Use the right amount of intelligence
Start fast for everyday work, then increase reasoning only when the problem truly earns the extra time.
Model choiceReasoning levelSpeed vs depth
A simple decision ladder
InstantRewrite, summarize, brainstorm
→Medium / HighCompare, judge, plan, analyse
→Extra High / ProDifficult, ambiguous, high-value work
✓Choose by the work in front of you
Higher reasoning is useful when the problem is hard—not when ChatGPT is missing information.
Medium or High
Use when the answer must compare evidence, balance constraints or reason across several steps.
Extra High or Pro
Reserve for unusually difficult analysis where the improvement is worth the additional time.
Model and reasoning are different controls
The model
Sets the underlying capability, speed and cost profile. In ordinary paid ChatGPT, model routing is often handled for you.
The reasoning level
Controls how much thinking that model applies to this response. More is not automatically better for simple work.
Context before horsepower
If the answer is weak because a file, fact or business constraint is missing, add the missing context. Increasing reasoning cannot invent reliable source material.
Important behavior behind the model picker
The options visible in ChatGPT depend on the account, plan and workspace controls around the conversation.
Automatic reasoning
On eligible paid plans, ChatGPT can increase reasoning automatically for a difficult request even when Instant is selected. This setting is managed separately from manually choosing a higher reasoning level.
More reasoning has a cost
Higher reasoning normally takes longer and can use more allowance. It is most useful for analysis, planning and diagnosis where the additional care can materially improve the result.
Availability is contextual
A managed workspace may restrict models or set defaults for members. Model names and access can change, so the model picker is the most reliable place to confirm what your account currently offers.
Models & Reasoning
Know when to turn reasoning up
Use task complexity—not anxiety—as the signal. The strongest setting is most valuable when there are real trade-offs, ambiguity or long chains of logic.
CapabilityLatencyConfidence
Match effort to stakes
📝Simple + reversibleInstant
📊Several factorsMedium or High
🧭AmbiguousHigh or Extra High
🏛️High stakesStrong model + review
Three questions before you change settings
How the family fits together
Model
Best fit
Remember
Sol
Complex professional work
Strongest general capability
Terra
Regular agentic work
Balances capability, speed and cost
Luna
Fast or high-volume work
Useful when the task is simpler
Always review consequential work
A confident answer can still be wrong. Check important numbers, assumptions and decisions against the source.
Why the same model behaves differently across products
Capability is shaped not only by the model, but also by the tools, context and permissions surrounding it.
Tools change the job
Where available, the model can work with web search, files, images, computer use and connected tools. Those capabilities explain why Chat, Work and Codex can feel different even when related models are involved.
Context limits still matter
Published API context specifications are not the same as the amount of information available inside every ChatGPT plan or product. Product limits and the quality of supplied context still apply.
Cost explains routing
Stronger models cost more to operate. Products may therefore use faster or lower-cost models for routine work and reserve the strongest capability for tasks where it creates meaningful value.