ChatGPT Got Five Upgrades in Four Days — Did...

Published: May 8, 2026 Impact: High — default model changed for 500M+ users; Gmail inbox access live; ads now matched to conversation history

Five Changes, Four Days, Zero Announcements

Between May 5 and May 8, 2026, OpenAI made five significant changes to ChatGPT. None were announced to users inside the product. All are live now.

OpenAI swapped the default ChatGPT model for every user on every tier — and most of the 500 million+ people who opened ChatGPT this week have no idea.

Here is what changed and when:

May 5: GPT-5.5 Instant replaces GPT-5.3 Instant as the default model across all tiers
May 5: Memory improvements roll out; Gmail inbox access live for Plus/Pro users who connected Gmail
May 6: Advanced Account Security features deploy (phishing-resistant sign-in, session controls, login alerts)
May 7: Ads confirmed expanding from US to Canada, Australia, and New Zealand — matched to conversation history
May 8: Excel and Google Sheets sidebar globally live for all users

The Model Change Most Users Missed

GPT-5.5 Instant is now the default ChatGPT model. GPT-5.3 Instant — the previous default — was replaced silently on May 5.

What changed in GPT-5.5 Instant, according to OpenAI's own data:

52.5% fewer hallucinated claims on high-stakes prompts (medicine, law, finance)
37.3% fewer inaccurate claims on flagged conversations
30.2% fewer words per response and 29.2% fewer lines
More natural conversational tone — less emoji, shorter answers

The improvements are real. The silence is not. OpenAI changed the default model for every free, Plus, Pro, Team, and Business user without a product notification. GPT-5.3 Instant remains accessible via model settings for three months, then retires. GPT-5.5 Instant is also deployed as chat-latest in the API, meaning developers who used chat-latest to stay on the current model were auto-migrated.

The Naming Confusion OpenAI Won't Explain

This is the critical watchdog point: GPT-5.5 Instant is not GPT-5.5.

OpenAI now has three active models with nearly identical names:

Model	What It Is	Pricing
GPT-5.5	Premium API model (April 23 launch)	$5/$30 per million tokens
GPT-5.5 Instant	Default ChatGPT consumer chatbot	Included in subscription tiers
GPT-5.4 mini	Budget/high-volume API model	Fraction of GPT-5.5 pricing

When we covered GPT-5.5's April 23 launch — a model with doubled API pricing and 82.7% agentic benchmark scores — that was a different product. GPT-5.5 Instant is the everyday chatbot model used by people who open chat.openai.com. They share a naming prefix. They are not the same model.

OpenAI does not explain the relationship between these models in its announcements. The naming makes it structurally difficult to compare what users on different tiers are actually receiving. That is not an accident.

What Is Reading Your Data Now

Gmail Memory

Plus and Pro users who previously connected Gmail now receive ChatGPT responses informed by their inbox. The feature works across memory improvements that also launched May 5.

Memory sources — including Gmail — are now visible and editable in Settings. You can see what ChatGPT has stored from your inbox, edit entries, or disconnect Gmail entirely.

What this costs you: ChatGPT can read your Gmail inbox to personalise responses. The trade-off is real and specific: your email content informs ChatGPT's memory, which persists across sessions. OpenAI states this is opt-in via the Gmail connection, but the default state for users who previously connected Gmail is active — the inbox access turns on with the memory update unless you manually disconnect.

If you connected Gmail months ago and forgot about it, your inbox is now a memory source. Check Settings → Connected Apps.

Verdict: For users who want a genuinely personalised assistant, this is a meaningful feature. For anyone with sensitive communications in Gmail — legal, financial, medical, personal — disconnecting is the conservative choice. The trade-off is real. Most users haven't been asked to re-confirm it.

Ad Targeting on Conversation History

Free and Go tier users now see ads matched to:

Current conversation topic
Past chat history
Past ad interactions within ChatGPT

This is not third-party cookie tracking. This is OpenAI using your conversation history as targeting data inside its own product. Plus, Pro, Business, Enterprise, and Education plans remain ad-free.

The geographic expansion — US to Canada, Australia, New Zealand — is also live.

If you are on the free tier and discussing health symptoms, legal questions, or financial decisions with ChatGPT, those conversations now inform the ads you see. OpenAI has not confirmed whether this data leaves the platform for third-party ad networks.

The Productivity Integrations

ChatGPT now runs as a sidebar inside Excel and Google Sheets — globally, as of May 8.

Use cases: building formulas, cleaning data, explaining workbook logic, generating pivot tables via chat. The sidebar is available without a paid subscription for basic use.

This is a direct move into Microsoft Copilot territory inside Microsoft's own product. The Excel integration is particularly notable — Microsoft owns OpenAI's largest investor position and also competes with it for enterprise productivity budgets.

Consumer Protection Q&A

Q: Did OpenAI tell me my default model changed? A: No. There was no in-product notification. The model selector in ChatGPT now shows GPT-5.5 Instant by default. If you want GPT-5.3 Instant, you can find it in model settings — for three months, then it is retired.

Q: Is GPT-5.5 Instant the same as the GPT-5.5 that launched April 23? A: No. GPT-5.5 (April 23) is a premium API model at $5/$30 per million tokens with 82.7% agentic benchmark performance. GPT-5.5 Instant is the consumer chatbot default. They share a name prefix. OpenAI does not explain the distinction in its communications.

Q: I didn't connect Gmail. Am I affected? A: No. Gmail memory requires an active Gmail connection. If you never connected Gmail, this change does not apply to you.

Q: I'm on the free tier. Are my conversations used for ad targeting now? A: Yes, if you are in the US, Canada, Australia, or New Zealand. Conversations, past chats, and past ad interactions are all used to match ads. This is stated in OpenAI's ad expansion announcement.

Q: What does the Advanced Account Security update actually change? A: Phishing-resistant sign-in (supports Yubico hardware security keys), tighter account recovery controls, shorter default session lengths, and login alerts. Automatic exclusion from model training is also toggled on as part of this update for accounts that enable the security features. Practically useful if you store sensitive data in ChatGPT memory.

What You Should Do

If you are on Plus or Pro:

Check Settings → Connected Apps — confirm whether Gmail is connected and decide if you want it active
Review your memory settings — you can now see and edit exactly what ChatGPT has stored from your inbox
GPT-5.3 Instant is still available in model settings; test GPT-5.5 Instant on your typical workloads before the 3-month retirement window closes
Enable Advanced Account Security if you store sensitive information in ChatGPT memory

If you are on the free tier:

Ads are now matched to your conversation history in the US, Canada, Australia, and New Zealand
If you discuss sensitive topics (health, legal, financial), consider this targeting mechanic before continuing those conversations in the free tier
Claude.ai's free tier and Gemini's free tier do not currently use conversation history for ad targeting

If you use the API:

chat-latest now routes to GPT-5.5 Instant. If you pinned chat-latest expecting stability, audit your costs — response length is 30% shorter, which affects output token usage
Pin a specific model version if you need consistent behavior across deployments

OneHuman Verdict

OpenAI: Five Changes, No Transparency ⬇ (Processes)

The GPT-5.5 Instant improvements are genuine — 52.5% fewer hallucinations on high-stakes prompts is a meaningful safety gain. The Gmail memory integration adds real personalisation value. The Sheets and Excel sidebars are practically useful.

The pattern around these improvements is the problem. A default model change affecting 500 million users, deployed with no notification. Gmail inbox access activated by association with a previous connection, not explicit re-consent. Ad targeting on conversation history rolled out through a blog post, not an in-product disclosure.

OpenAI's naming structure — GPT-5.5, GPT-5.5 Instant, GPT-5.4 mini — makes comparison harder, not easier. That is a deliberate product decision, not an oversight.

The features are improving. The transparency is not. For free tier users especially: you are now an advertising target, and your conversation history is the targeting data.

Sources:

Introducing GPT-5.5 Instant — OpenAI Official — May 5, 2026
Testing Ads in ChatGPT — OpenAI Official — May 2026
OpenAI releases GPT-5.5 Instant — TechCrunch — May 5, 2026
OpenAI updates ChatGPT Instant with GPT 5.5 — Axios — May 5, 2026
GPT-5.5 Instant makes ChatGPT smarter with fewer emoji — 9to5Mac — May 5, 2026
Verified by OneHuman: May 8, 2026

ChatGPT Got Five Upgrades in Four Days — Did You Notice?