ChatGPT Buries Its Model Selector: Enter the Effort Slider

On August 6, 2026, OpenAI quietly revamped the main interface of ChatGPT: the small dropdown menu listing available models (GPT-4o, o3, GPT-5, GPT-5.6…) disappears. In its place: an effort slider — fast, balanced, max thinking. Now, GPT-5.6 Sol makes the call under the hood, deciding how much reasoning to apply to each request.

A UX decision affecting one billion monthly active users. And it follows a similar shift made three weeks earlier with Claude.

What Changes Practically

• No more stress of “which model do I choose for this question?” GPT-5.6 Sol decides.
• The slider shows three positions: Fast (Luna under the hood), Balanced (Terra), Max thinking (Sol full reasoning).
• Plus/Pro users see a finer slider with Medium / High / Extra High effort.
• The old model selector hasn't disappeared — it's moved to Settings. One more click.

The Logic: Relieving Users of the Choice

OpenAI measured it: most users don't understand the difference between models, and many get stuck with the wrong choice for months. Result: 68% fewer errors in complex use cases since the shift, according to internal data from ExplainX.

Sam Altman has advocated for this approach for months: “the model should adapt to the human, not the other way around”. The philosophy extends to automatic routing — the model chooses the right model.

This is the moment when ChatGPT moves from “selling you configurations” to “selling you an experience.”

Three Tiers, Economic Continuity

Under the hood, the GPT-5.6 range remains segmented by price:

• Sol: flagship, max reasoning, $5/$25 per M tokens
• Terra: performance/cost balance, $2/$12
• Luna: the cheapest, $0.20/$1.20 (see our article on the price war)

The effort slider doesn't hide this segmentation; it automates it. Massive “fast” usage consumes Luna, cheaper for OpenAI. “Max” usage consumes Sol, more profitable. The arbitration algorithm optimizes margin and user perception simultaneously.

The Context: The UX War

Anthropic made the same choice three weeks ago with Claude, adding a finer effort ladder to Claude Opus 5. Google Gemini has been doing it since Gemini 3.1 with its “Deep Think” mode. It's now an emerging standard among the three major labs.

The unified message sent to users: stop choosing a model, choose an intention (fast, precise, exhaustive). It's a UX convergence that masks a fierce competition on inference costs.

What It Changes for Developers

Nothing immediately: the API still exposes the models distinctly (Luna, Terra, Sol). The effort slider is a ChatGPT interface artifact, not an API contract.

But in the medium term, we can imagine an auto-arbitration API endpoint where OpenAI decides the model. This already exists with OpenRouter (automatic routing), likely the next step for OpenAI for its large enterprise clients.

The Nuance

Power users lament the lost precision. It's now impossible to know exactly which model handled a given request — thus impossible to scientifically compare responses. On specialized tasks (legal, medical, code), several users on Reddit report regressions since the shift when they explicitly used o3 or GPT-5.5 before.

Key Takeaway

ChatGPT continues its transformation into a transparent consumer product, at the expense of fine technical control. For the 99% of users who don't know the difference between Sol and Luna, it's progress. For the 1% of power users who meticulously optimized their models, it's a regression. OpenAI bets that the first group generates 99% of the value — a coherent economic bet.

The next logical step: complete disappearance of the model concept on the user side. You only interact with “ChatGPT.” What's running behind becomes an invisible commodity.