ChatGPT's Free-Tier Model Changed on August 6. What We Know and Don't.

On August 6, 2026, ChatGPT changed the model serving its free and Go tiers. GPT-5.5 Instant -- the model that had defined what most ChatGPT users experienced -- was replaced by GPT-5.6 Luna as the default for unpaid accounts.

Most businesses building AI visibility strategies don't know this happened. The ones that do are asking a reasonable question: does Luna change the citation picture in a way that requires adjusting the fix plan?

The short answer: the fix plan doesn't change because the structural lever -- data partnerships -- operates below the model tier. But the longer answer matters, and it involves being direct about what we don't yet know.

The model that most of your customers use just changed

The tier structure currently in place:

- **Free/Go tier**: GPT-5.6 Luna (new default as of August 6, 2026) - **Plus tier**: GPT-5.6 Sol - **Pro/Business tiers**: Sol or Terra, selectable

Luna is a new default for the largest share of ChatGPT users. When a homeowner on the free tier asks "who are the best electricians near me," they are now getting a Luna response, not a GPT-5.5 Instant response.

Our methodology research first documented the Sol/Terra/Luna tier structure in Scout session 85 (July 23, 2026). A critical update followed in Scout session 105 (August 12, 2026) confirming the free-tier switch had gone live. The session 105 rec flagged the transition explicitly: as of August 6, 2026, the free/Go tier default changed from GPT-5.5 Instant to GPT-5.6 Luna.

What the old tier split revealed

The prior tier structure -- Instant for free users, Thinking for Pro users -- produced a measurable citation gap. Our July 17, 2026 edge-case investigation (Scout session 80) documented numbers that made the gap concrete: GPT-5.5 Thinking showed a 47.2% brand site citation rate in web search-enabled queries. GPT-5.5 Instant ran at approximately 6%. That is an 8:1 difference in how frequently the model pulls in web-sourced business citations when answering queries.

The interpretation for businesses: a user on the Pro tier asking a local recommendation query was dramatically more likely to see web-cited results than a user on the free tier. Free-tier responses leaned on training data rather than live web retrieval. That made Foursquare, Yelp, and directory partnerships -- which feed ChatGPT at the data layer before web retrieval -- the higher-leverage action for any business whose customers skew free-tier.

What we don't know about Luna

Here is where we need to be direct about the limits of current research: GPT-5.6 Luna's web citation behavior for local services queries has not been empirically measured as of our August 12, 2026 vault update.

It is tempting to look up Luna's benchmark scores -- accuracy metrics, reasoning tests -- and draw inferences about how frequently it will cite local business websites when answering recommendation queries. That inference is not sound. Our July 17 investigation flagged this problem when analyzing the Sol/Terra gap: in-context document retrieval benchmarks test how well a model uses documents passed directly to it, not how frequently the model performs web retrieval and cites sources. These are different behaviors. A model with strong benchmark scores may still behave like GPT-5.5 Instant in citation terms -- pulling from training data rather than live web retrieval.

We don't have 50+ prompt studies on Luna's local services citation behavior. Until that data exists, the session 105 rec is conservative: maintain third-party directory presence as primary fix lever for free-tier users. That is not because Luna is expected to underperform. It is because we have not measured it.

The session 105 rec logged a specific reopen trigger: any published study measuring GPT-5.6 Luna citation behavior for local services queries at 50 or more prompts. When that study surfaces, we will update the guidance.

What hasn't changed: the data partnership layer

The reason Luna uncertainty doesn't change the fix sequence is architectural.

ChatGPT's responses to local service queries are assembled through data partnerships that operate before model-level web retrieval. Foursquare (confirmed December 2024) supplies the POI database that populates the majority of ChatGPT local results -- a Yext study of 6.8 million ChatGPT citations traced 60-70% of local business results back to Foursquare. Yelp (July 2026) licensed 330 million reviews and 8 million+ business listings directly to OpenAI via structured API. Thumbtack (October 2025) integrates home services professionals directly into standard ChatGPT responses.

These partnerships are data feeds. They are not filtered by tier. A free-tier user on Luna and a Pro user on Terra asking the same local services query draw from the same Foursquare database, the same Yelp data feed, the same Thumbtack integration. The model tier affects how ChatGPT reasons about and presents that data -- and may affect how frequently it reaches for live web retrieval on top of it -- but it does not affect whether your business is in the candidate pool.

Our August 2, 2026 methodology update (Scout session 95) established this framing: for ChatGPT gaps, third-party directory presence and Foursquare/Yelp data quality are the effective levers. The August 6 Luna change does not alter that priority order. The data partnerships are the access point to ChatGPT recommendations regardless of tier.

Two things worth tracking

The Luna citation behavior question will get answered. AEO researchers and citation tracking platforms will publish spot tests. When that data arrives, we will have an empirical Luna baseline the way we have Instant and Thinking baselines.

Until then, two dynamics are worth watching.

First, the Sol/Luna gap for Plus versus free users. The prior Thinking/Instant gap was approximately 8:1 in citation rate. Whether Sol/Luna produces a comparable split is structurally plausible -- Luna is a newer, capable model but still the free tier -- but it is not documented. Businesses with customers concentrated on Plus subscriptions may see different AI visibility patterns than those whose customers mostly use the free tier.

Second, whether Luna changes organic web retrieval frequency relative to Instant. If Luna retrieves from the web more aggressively than Instant did, organic website signals -- schema, content structure, freshness -- would gain more weight for free-tier users than they carried under the Instant baseline. We will update the fix plan accordingly when that data is available.

The fix sequence stands

For businesses building ChatGPT visibility right now, the August 6 Luna change does not alter the action sequence. Foursquare claim, Yelp profile completeness and review volume, and Thumbtack listing for home services remain the structural levers that feed the data layer across all tiers. These are not proxy signals -- they are direct inputs to how ChatGPT assembles local recommendation responses.

The Luna question is real, and the absence of measured data is the honest position to hold. Filling that gap with benchmark extrapolation produces guidance that looks confident and may be wrong. The data partnerships approach holds because it works at the layer that doesn't depend on which model is running.

If you want to see where your business currently stands across ChatGPT, Perplexity, Gemini, and Google -- including which data feeds you are and are not in -- Signal Check at sourcepull.ca runs a live platform-by-platform audit. The gaps it surfaces are the ones that affect every ChatGPT user, regardless of which tier they are on.