OpenAI updated GPT-5.6 Sol in ChatGPT only — the Sol behind your API and Codex calls did not change

On August 6, 2026 OpenAI shipped a chat-tuned GPT-5.6 Sol to Plus and Pro, made GPT-5.6 Luna the free default with unlimited text chats, and added an effort slider. Buried in the availability section: this build ships to the ChatGPT Chat experience only, and the Sol powering Work and Codex is unchanged. That single sentence is the whole story for builders.

OpenAI shipped an updated GPT-5.6 Sol to ChatGPT on August 6, 2026, and expanded free-tier access to GPT-5.6 Luna. (Source: OpenAI, 2026-08-06)

Key facts:

The updated GPT-5.6 Sol is available to Plus and Pro users starting today (August 6, 2026).
A new effort slider ships with it, on web, mobile, and desktop — you pick how much thought ChatGPT puts into each answer.
For Plus and Pro, the same model now powers both Instant responses and deeper reasoning. There is no longer a separate Instant model underneath.
GPT-5.6 Luna becomes the default model for Free and Go users this week.
Unlimited text chats and a new Think button reach Free and Go users starting next week, subject to abuse guardrails. Limits still apply to file uploads, images, and other tools.
On an internal evaluation of financial, medical, and legal prompts requiring factual detail, responses containing at least one factual error were about 62% less common with GPT-5.6 Luna and 68% less common with GPT-5.6 Sol than with GPT-5.5 Instant.
This build of Sol is limited to the Chat experience in ChatGPT. OpenAI states plainly: “The version of GPT‑5.6 Sol that powers Work and Codex is not changing as part of this release.”
OpenAI cites 1 billion people using ChatGPT weekly.

ChatGPT Free interface on a starfield background showing a Get smarter answers dialog with a Think button, an Upgrade to Plus link, and Upgrade to Plus and Turn on buttons. — The Think button and its upsell dialog, as shipped to Free users. The free tier now gets a manual reasoning toggle rather than automatic routing. (Image: OpenAI, 2026-08-06)

What this means if you’re building on GPT-5.6

1. ChatGPT is no longer a valid preview of your API behaviour. This was always a soft rule; today it became an explicit product fact. The Sol you talk to in ChatGPT Chat has been tuned for everyday conversation — tighter formatting, less hedging, fewer volunteered details. The Sol your API calls and your Codex runs hit is the previous build. If your prompt-engineering loop is “try it in ChatGPT, then paste into the app,” that loop now silently compares two different models. Test in the surface you ship on.

Screenshot of the Safety and availability section of OpenAI's announcement, stating that Plus and Pro users get the updated GPT-5.6 Sol and slider starting today, that GPT-5.6 Luna becomes the Free and Go default this week with unlimited text chats next week, and that the version of GPT-5.6 Sol powering Work and Codex is not changing. — The availability paragraph, verbatim. The last sentence is the one that matters if you ship on the API or run Codex. (Screenshot of OpenAI's announcement, 2026-08-06)

2. The “62% / 68% fewer factual errors” number is narrower than it reads. It is an internal OpenAI evaluation, on financial, medical, and legal prompts that require factual detail, measuring the share of responses containing at least one factual error, against GPT-5.5 Instant as the baseline. It is not a claim against GPT-5.6 Sol’s previous build, and there is no public eval set behind it. Treat it as a directional statement about the chat build and keep your own regression set.

3. Collapsing Instant and Thinking into one model is the structural change. For Plus and Pro, one model now serves both quick answers and extended reasoning, with the slider controlling effort rather than swapping models. That is the same shape as the reasoning-effort parameter developers already use through the API — the consumer product is converging on the developer control surface. If you have been treating “fast tier” and “thinking tier” as two products with two prompt styles, the direction of travel is one model plus an effort dial.

4. Free-tier unlimited text chats change the calculus for consumer-facing wrappers. If your product’s pitch is “ChatGPT-quality answers without a subscription,” the free tier now offers unlimited text chats on GPT-5.6 Luna plus a manual Think button starting next week. The remaining gaps are file uploads, images, and other tool limits — which is where a wrapper still has room, and where it no longer does.

5. Nothing here is an API pricing or model-ID change. No new model IDs, no rate-limit changes, no deprecations were announced. If you are choosing between Sol, Terra, and Luna for a workload, the earlier price-performance positioning still stands.

Sources: OpenAI, “Improving GPT-5.6 Sol in ChatGPT”, 2026-08-06 · OpenAI, “Advancing the price-performance frontier with GPT-5.6”, 2026-07-30 · OpenAI News