OpenAI Retiring GPT-4.5 and o3 from ChatGPT: What Free Users Need to Know (2026)

If you logged into ChatGPT this week and noticed fewer options in the model selector, you're not imagining things. OpenAI quietly announced another round of retirements through its official Model Release Notes — and the headline is two models that many users still rely on: GPT-4.5 and o3. The announcement dropped with minimal fanfare just days ago, but the implications are real, especially for the tens of millions of people using ChatGPT on the free tier.

This is not OpenAI's first rodeo with model retirements in 2026. Back in February, the company removed GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini from ChatGPT in a single sweep — a move that triggered a wave of backlash from users who had developed strong attachments to those models. Now, just four months later, the cleanup continues.

What ChatGPT models is OpenAI retiring in mid-2026?

The two models confirmed for retirement are GPT-4.5 and o3. Both announcements appear in the same release note update, suggesting OpenAI is continuing its deliberate effort to consolidate ChatGPT around the newer GPT-5.5 family.

GPT-4.5 was quietly positioned as a bridge between GPT-4o and the full GPT-5 launch. It offered improved instruction-following and more nuanced responses, but it was available only through the model settings selector for paid subscribers — free users never had default access to it. As of the announcement, GPT-4.5 is already flagged as paid-only access, making its June 27 removal less disruptive than it might sound.

OpenAI o3 is the more consequential retirement, especially for users who value reasoning-focused AI. The o3 model was OpenAI's flagship reasoning system — capable of multi-step logical analysis, mathematics, and complex coding tasks that standard GPT models struggle with. It was available to free users, but with severe limits: according to independent testing documented by PanelsAI, free-tier users received roughly a few o3 messages per week before hitting their quota.

GPT-4.5's retirement also marks a symbolic milestone. TechRadar described it as "quietly retiring the final GPT-4 model from ChatGPT" — a complete close of the chapter that defined early mainstream AI adoption. The GPT-4 era, which began in 2023 and changed how millions of people work, will be fully gone from ChatGPT's interface by the end of June.

When exactly are the retirement dates?

OpenAI has confirmed two specific retirement windows, both announced through the ChatGPT Release Notes page:

Model	Retirement Date	Sunset Period	Who's Affected
GPT-4.5	June 27, 2026 26 days	30-day sunset	Paid subscribers with model selector access
OpenAI o3	August 26, 2026 ~87 days	90-day sunset	Paid users (heavy) + free users (limited weekly quota)
GPT-4.5 (API)	No change	—	Developers unaffected — API access continues
o3 (API)	No change	—	Developers unaffected — API access continues

One important clarification from OpenAI's own documentation: "These changes apply to ChatGPT only; there are no changes to the API." Developers building applications on top of o3 or GPT-4.5 via the API do not need to change anything on these dates.

What does ChatGPT's 2026 model lineup actually look like now?

If you're confused by how quickly OpenAI's model roster has churned this year, you're not alone. Here's the current landscape across ChatGPT tiers as of June 2026:

• Free ($0/month) — GPT-5.5 is the default model, with caps of approximately 10 flagship messages per 5-hour window. After hitting limits, the system downgrades to a lighter version. Access to o3 reasoning is limited to a few queries per week until August 26.
• Go ($8/month) — OpenAI's entry-level paid tier, launched globally in January 2026. More generous message limits than Free with GPT-5.5.
• Plus ($20/month) — The workhorse tier. GPT-5.5 access at higher limits, including Thinking Mode, voice, and code interpreter tools. Codex included.
• Pro ($100/month) — Launched April 9, 2026. Up to 200 o3 messages per day and 5× usage limits compared to Plus before the retirement. After August 26, Pro shifts fully to GPT-5.5-based reasoning.
• Business ($20/seat/month) — Team collaboration features, admin tools, and higher throughput.
• Enterprise (custom) — Pooled usage, SCIM provisioning, audit logs, and dedicated capacity.

For a detailed breakdown of how each plan stacks up on actual usage limits, see our free vs paid AI comparison guide — which we keep updated as OpenAI changes its tier structure.

What do free users actually lose when o3 retires?

This is the question that matters most for the majority of ChatGPT users. The honest answer: less than the headlines suggest, but not nothing.

On the GPT-4.5 side: free users lose nothing. The model was already behind a paid-tier gate. The June 27 retirement is primarily a cleanup of the model selector for Plus and Pro subscribers who may have been manually selecting it.

On the o3 side: free users do lose their small weekly allocation of reasoning-mode queries. For casual users who never knew they had o3 access, this is invisible. For those who deliberately used o3 for the occasional hard math problem or multi-step reasoning task — and hit their quota every single week — this is a real downgrade.

The silver lining is genuine. OpenAI's Help Center Free Tier FAQ now states that "Free tier users have access to a large range of capabilities with GPT-5.5" — and GPT-5.5 is meaningfully more capable than o3 at most practical tasks. For general conversation, writing, analysis, and coding assistance, GPT-5.5 outperforms o3 in the contexts most users actually encounter. o3's advantage was specifically in structured reasoning and benchmark-style problem solving.

If you use ChatGPT free for research, writing help, coding, or everyday questions, your experience will be the same or better after August 26. If you were using those few weekly o3 queries for hard logic problems or advanced mathematics, you'll need to either upgrade to a paid plan or explore free AI tools that still offer reasoning access.

Why is OpenAI retiring models so quickly in 2026?

The acceleration of OpenAI's retirement cycle is not accidental. The company is executing a deliberate strategy to consolidate its product around the GPT-5.5 family — and there are clear business incentives behind it.

Running multiple large language models in production simultaneously is enormously expensive. Every token generated by GPT-4.5 or o3 in ChatGPT consumes server capacity that OpenAI would rather allocate to its newest, most commercially significant models. By retiring older models from the chat interface, OpenAI reduces infrastructure complexity while nudging users toward GPT-5.5 — which it can serve more efficiently at scale.

There's also a strategic signal in the timing. The wave of retirements in 2026 began in February with the GPT-4 family, and now continues with GPT-4.5 and o3. A clear message is forming: ChatGPT is a GPT-5.5 product now. The older model names are being erased from the consumer interface even as they remain available through the API for developers who need continuity.

The Verge noted in April that the broader industry trend is toward what it called "the AI money squeeze" — a period where AI companies that have been giving away capable models for free are moving toward monetization models that require paying for the best capabilities. Model retirements are part of that same broader shift: if you want the old reasoning models, you need a Pro plan. If you're on Free, you get what OpenAI decides the default should be.

You can track all current free tier limits and changes across every major provider on our free tier tracker — updated whenever announcements like this drop.

Should free users upgrade before the retirement dates?

For the vast majority of ChatGPT's free users: no. The retirements are mostly invisible to everyday usage.

However, there are specific cases where upgrading before August 26 makes sense:

• You regularly hit your o3 weekly limit — If you've been maxing out your free o3 quota every week for complex reasoning tasks, those queries go away entirely for free users in August. The Go plan at $8/month or Plus at $20/month give you meaningfully higher allocations now while o3 is still running.
• You have a specific workflow that depends on o3's benchmark reasoning — Some tasks like competition-level math, formal logic proofs, or structured code debugging benefited from o3's specific architecture. GPT-5.5 handles most of this well, but if you've tuned a workflow to o3 specifically, test GPT-5.5 Thinking Mode before the August deadline.
• You're a developer building on ChatGPT — API access is unaffected by both retirements, so no action needed from builders using o3 or GPT-4.5 via the API directly.
• You want to archive GPT-4.5 outputs for reference — If you have ongoing threads or use cases that specifically leverage GPT-4.5's response style, June 27 is your hard cutoff. Export anything you need before that date.

For alternatives beyond ChatGPT, our model news section tracks reasoning-capable tools across Google, Anthropic, and open-source providers — several of which maintain free reasoning access that OpenAI is now scaling back.