When does GPT-4.5 retire?

OpenAI is retiring GPT-4.5 from ChatGPT on June 27, 2026, following a 30-day sunset period that began in late May 2026. GPT-4.5 will no longer be selectable in the ChatGPT model picker as of that date. On the same OpenAI deprecation announcement, o3 (the previous-generation reasoning model) is scheduled for retirement from ChatGPT on August 26, 2026 following a 90-day sunset. The pattern reflects OpenAI's accelerated deprecation cadence as GPT-5, GPT-5.5, and GPT-5.5 Instant become the default models across Free, Plus, Pro, and Enterprise tiers. API access timelines for GPT-4.5 may differ from ChatGPT and were not fully detailed in the public announcement; developers using the API should consult the OpenAI deprecation documentation directly.

What should I use instead of GPT-4.5?

For most ChatGPT users, GPT-5.5 Instant is the default replacement and is generally faster and more capable than GPT-4.5 for everyday tasks. For longer or more analytically demanding tasks, GPT-5 (the slower, more deliberate variant) is appropriate. For reasoning-heavy workflows that previously used o3, the replacement path is GPT-5 with its built-in extended reasoning, since dedicated o-series reasoning models have been merged into the GPT-5 family. For coding, Codex (the agentic coding product) plus GPT-5.5 is the standard combination. For developers using GPT-4.5 via the API, the migration is similar — GPT-5 family models are the default replacement; review the OpenAI deprecation documentation for the specific API model IDs and their recommended replacements.

Why is OpenAI retiring GPT-4.5 so quickly?

GPT-4.5 was a transitional model — it launched in February 2025 as 'OpenAI's largest and most knowledgeable model' but was always positioned as a step on the path to GPT-5 rather than a long-term flagship. With GPT-5 (Aug 2025), GPT-5.5 (early 2026), and GPT-5.5 Instant (mid-2026) all released and widely deployed, the gap between GPT-4.5 and the current default has grown significantly. Maintaining GPT-4.5 in production has compute and operational cost; retiring it lets OpenAI focus resources on the GPT-5 family. The 30-day sunset is faster than the historical 12-month deprecation windows OpenAI used in 2023-2024 — this reflects the accelerated cadence of the 2026 frontier-model market, where models become obsolete faster than ever before.

How do I migrate ChatGPT custom GPTs, projects, and saved chats from GPT-4.5?

ChatGPT will automatically migrate users to a default model after June 27, 2026 — typically GPT-5.5 Instant for everyday use and GPT-5 for longer/analytical tasks. Saved chats remain accessible; the model selector for new turns in an existing conversation falls back to the current default. Custom GPTs that were configured with GPT-4.5 will fall back to the default model; in most cases this is transparent but you may want to re-test custom GPTs that rely on GPT-4.5-specific behaviors. Projects (the ChatGPT feature for grouping chats with shared context) similarly migrate to the default model. For API users, applications hard-coded to gpt-4.5 model IDs will start failing after the API deprecation date; update model IDs in your code before then and re-run regression tests against the GPT-5 replacement model.

Quick Answer

GPT-4.5 Retires June 27, 2026: What To Use Instead

Published: June 26, 2026

GPT-4.5 Retires June 27, 2026: What To Use Instead

OpenAI is retiring GPT-4.5 from ChatGPT on June 27, 2026 after a 30-day sunset. o3 follows on August 26, 2026 after a 90-day sunset. GPT-5.5 Instant is the everyday default; GPT-5 is the longer-context, more deliberate variant. For developers using GPT-4.5 via the API, the migration is similar — but check the specific deprecation page for API model ID timelines and replacements.

Last verified: June 26, 2026 (one day before retirement).

TL;DR

GPT-4.5 retirement in ChatGPT: June 27, 2026 (30-day sunset ends)
o3 retirement in ChatGPT: August 26, 2026 (90-day sunset ends)
Default replacement: GPT-5.5 Instant for everyday; GPT-5 for analytical/long-context
For coding agents: Codex + GPT-5.5
API users: check OpenAI deprecation docs for API model ID timelines
Saved chats and custom GPTs: auto-fall-back to current default; re-test for any GPT-4.5-specific behaviors

The retirement, exactly

OpenAI announced the GPT-4.5 ChatGPT retirement in late May 2026 with a 30-day sunset window. The deprecation page lists:

Model	Surface	Retirement Date	Sunset Length
GPT-4.5	ChatGPT	June 27, 2026	30 days
o3	ChatGPT	August 26, 2026	90 days
Older GPT-image models	API	Various, mid-2026	per model
v1/prompts API + reusable prompt objects	API	Nov 30, 2026	extended

Sunset behavior in ChatGPT:

During the sunset window: the model is still selectable in the model picker; a deprecation banner is shown
After retirement: the model is removed from the picker; existing chats fall back to the current default model for new turns

What replaces GPT-4.5

Everyday use → GPT-5.5 Instant

GPT-5.5 Instant is the current default for ChatGPT across Free, Plus, Pro, and Enterprise tiers. It’s faster than GPT-4.5, has better instruction following, and handles conversational quality (context handling, follow-up coherence, local and product recommendations) better. For 80%+ of GPT-4.5 use cases, this is the drop-in replacement.

OpenAI updated GPT-5.5 Instant on June 24, 2026 to further improve conversational quality — context handling, instruction following, decision/advice/planning/research/shopping quality. The update is part of the explicit upgrade path for retiring GPT-4.5 users.

Long-context / analytical → GPT-5

GPT-5 is the slower, more deliberate variant in the GPT-5 family. It has stronger reasoning, longer context handling, and better performance on complex analytical tasks. For tasks that previously used GPT-4.5 because of its knowledge breadth, GPT-5 is the better target.

Reasoning workflows previously on o3 → GPT-5

The o-series reasoning models have been merged into the GPT-5 family — reasoning is now an integrated capability rather than a separate model line. Workflows that previously selected o3 explicitly should default to GPT-5; the model handles extended reasoning internally without requiring a separate model selection.

For agentic, multi-step, or coding tasks: pair GPT-5 with Codex (OpenAI’s agentic coding product). This is the recommended combination for long-running work.

Coding workflows → Codex + GPT-5.5 / GPT-5

For ChatGPT-based coding: Codex (agentic mode) is the primary tool, backed by GPT-5.5 for fast iterations and GPT-5 for harder problems. Codex Security (released June 22-23, 2026 as part of the Daybreak expansion) is the security-focused variant.

API migration

The ChatGPT timeline and the API timeline are separate. Public reporting confirmed the ChatGPT dates; the API model deprecation timelines are listed on OpenAI’s platform deprecation documentation (consult directly for current information).

Migration steps for API users

Audit your code for model IDs. Search for gpt-4.5, gpt-4.5-preview, gpt-4.5-turbo (or similar) across your codebase, infrastructure-as-code, and config files.
Pick the replacement model. GPT-5 family is the default replacement; choose gpt-5 for general use, gpt-5-mini for cost-optimized cases (when available), or domain-specialized variants like GPT-5.5-Cyber for security workflows.
Re-test prompts. GPT-5 family models often need lighter prompts than GPT-4.5; over-elaborate instructions can degrade output. Re-test key prompts and simplify where possible.
Update token budgets. Pricing differs between GPT-4.5 and GPT-5 family; recompute your per-request cost estimates and update budgets.
Re-evaluate evals. Run your offline evals against the new model and tune thresholds where needed.
Deploy with feature flags. Roll out gradually with A/B testing if possible; watch for regression in quality metrics specific to your application.

Why the sunset is so fast

OpenAI’s 2023-2024 deprecation cadence was approximately 12 months. The current 30-day cadence reflects several pressures:

1. Compute reallocation

Maintaining GPT-4.5 inference capacity costs real money. Repurposing that compute to GPT-5 family inference improves margins and serves more users with the current default.

2. The 2026 model release cadence

Frontier model releases are happening monthly. OpenAI has released GPT-5 (Aug 2025), GPT-5.5 (early 2026), and GPT-5.5 Instant (mid-2026), and is reportedly working on GPT-5.6 Pro. Keeping every prior generation alive is operationally infeasible.

3. Product clarity

When ChatGPT showed 6+ models in the picker, user choice complexity hurt the product. OpenAI has been simplifying the model picker; retiring GPT-4.5 is part of that.

4. Competitive pressure

Anthropic, Google, xAI, and others retire models on similarly fast cadences. The frontier-model market in 2026 is structurally less stable than 2023-2024.

What to actually do today (June 26, 2026)

If you’re a ChatGPT user:

Tomorrow (June 27), GPT-4.5 disappears from the picker
You don’t need to do anything; ChatGPT will default you to GPT-5.5 Instant (or GPT-5 for longer threads)
If you have specific GPT-4.5 prompts that work well, save them as text and re-test on GPT-5.5 Instant before relying on them in production work

If you’re a developer using GPT-4.5 via the API:

Check the OpenAI deprecation page for API-specific dates and model IDs
Update model IDs in your code to GPT-5 family before the API deprecation date
Re-run regression tests against the new model
If you have specific prompts tuned to GPT-4.5, simplify them — GPT-5 family usually performs better with cleaner prompts

If you’re an enterprise running custom GPTs or Projects:

Audit any custom GPTs that explicitly require GPT-4.5; re-test them under the new default
Notify users of any expected behavior changes
Update internal documentation to remove GPT-4.5 references

How this fits the broader 2026 model landscape

Provider	Current Flagship	Notable Recent Retirement
OpenAI	GPT-5.5 (with GPT-5 for hard tasks)	GPT-4.5 → June 27, 2026; o3 → Aug 26, 2026
Anthropic	Claude Fable 5 (June 9, 2026)	Multiple deprecations June 15, 2026
Google	Gemini 2.5 Pro with Deep Think (June 23, 2026)	Earlier 2.0 line deprecated through 2026
xAI	Grok 4 family	Older Grok versions deprecated
Mistral	Mistral Large 2 + Mixtral updates	Older Mixtral versions deprecated

The pattern is industry-wide: frontier model providers are sunsetting prior-generation models on 30-90 day windows. This is the new normal; expect the next wave of deprecations within 6 months.

Bottom line

GPT-4.5 retires from ChatGPT on June 27, 2026 (one day after this article’s publication). The migration path is simple for most users (GPT-5.5 Instant is the default replacement) and more involved for API users (audit model IDs, re-test, update budgets). The fast sunset cadence is the new normal in the 2026 AI market — build your deployment processes to expect 30-90 day deprecation windows for any model you depend on.

The bigger lesson: don’t hard-code model identity in user-facing experiences or developer documentation. Model names will change again, probably within months. Build adapter layers that map your application’s notion of “default fast model” and “default deliberate model” to current OpenAI (or Anthropic, or Google) model IDs, and you’ll absorb the next deprecation with minimal pain.