OpenAI is retiring GPT-4.5 from ChatGPT on June 27, 2026 — 26 days from today. If you rely on GPT-4.5 via the API or in a ChatGPT workflow, you need to act now. Here’s exactly what’s changing, what to migrate to, and when.
What OpenAI Is Sunsetting
OpenAI announced a two-model sunset in June 2026:
- GPT-4.5: Removed from ChatGPT and deprecated from the API on June 27, 2026
- o3 (reasoning model): Deprecated on August 26, 2026
GPT-4.5 was released in February 2026 as OpenAI’s “most knowledgeable model” at the time — a large, expensive, conversationally capable model trained on massive data. It was never the fastest or cheapest option, but many developers and power users chose it for its broad world knowledge and smooth conversational quality.
The retirement comes just four months after launch, which is fast even by OpenAI’s standards. The reason: GPT-5 and GPT-5.3-Codex have fully superseded it on every benchmark that matters.
What to Use Instead
OpenAI’s recommendation is straightforward:
| If you used GPT-4.5 for… | Switch to |
|---|---|
| General conversation, research, writing | GPT-5 (gpt-5) |
| Coding tasks, agentic workflows | GPT-5.3-Codex (gpt-5-3-codex) |
| Cost-sensitive applications | GPT-4.1-mini or GPT-4.1-nano |
| Complex multi-step reasoning | o4 (replacing o3 before August sunset) |
GPT-5 is faster, cheaper per token, and scores higher on most benchmarks than GPT-4.5 did at launch. There’s no reason to stick with GPT-4.5 if you have access to GPT-5 — the migration is essentially an upgrade.
API Migration Steps
If you’re calling GPT-4.5 via the API, here’s what to do before June 27:
- Find your calls: Search your codebase for
gpt-4.5,gpt-4.5-turbo, orgpt-4.5-previewmodel strings. - Swap the model string: Replace with
gpt-5for general use orgpt-5-3-codexfor coding/agentic tasks. - Test outputs: GPT-5 is more capable but may respond differently to the same prompts. Run your test suite against the new model — don’t assume identical behavior.
- Update your prompts if needed: GPT-5 follows instructions more precisely, which can mean prompts that worked around GPT-4.5’s limitations now behave differently.
- Check token limits: GPT-5 has a larger default context window. If you were chunking inputs for GPT-4.5, you may be able to simplify your pipeline.
What Happens If You Don’t Migrate?
After June 27, API calls to gpt-4.5 model strings will return an error. ChatGPT users on plans that defaulted to GPT-4.5 will be automatically shifted to GPT-5 — no action required for regular ChatGPT users. The disruption is only for developers and businesses with hardcoded API integrations.
The Broader Pattern: OpenAI’s Accelerating Model Cadence
GPT-4.5 lived for four months. GPT-4 lasted about 18 months before being sidelined. This compression is intentional — OpenAI is shipping models faster than ever and retiring older ones to reduce the maintenance burden and push the entire ecosystem toward newer, better models.
The implication for developers: don’t hardcode model strings. Use an environment variable or configuration value for model names so you can update without a code deploy. OpenAI’s rapid release cadence means the model you build on today might be deprecated within a year.
o3 Sunset: What You Need to Know
OpenAI’s o3 reasoning model follows GPT-4.5 out the door on August 26, 2026. o3 was OpenAI’s flagship reasoning model for much of late 2025, capable of extended chain-of-thought reasoning on complex math, science, and coding problems.
Its replacement is o4, which OpenAI says is faster, cheaper, and more accurate on reasoning benchmarks. If you’re using o3 for complex reasoning tasks, the migration window is longer — you have until August 26 — but the steps are the same: swap o3 for o4 and test your outputs.
Key Dates
- June 27, 2026: GPT-4.5 removed from ChatGPT and API
- August 26, 2026: o3 reasoning model deprecated from API
For most users, the transition to GPT-5 is a straightforward upgrade. But if you’re running production applications on GPT-4.5 via the API, 26 days is a tight window. Start the migration today.