GPT-4o & GPT-4.1 API Deprecation: Migrate to gpt-6-luna
GPT-4o API deprecation has a January 14, 2027 deadline. See affected model IDs, the earlier fine-tuning cutoff and the GPT-6 Luna migration checks.
OpenAI's October 14 notice schedules five GPT-4o and GPT-4.1 API snapshots for removal on January 14, 2027; its recommended replacement for each is gpt-6-luna. The gpt-4o api deprecation is a migration deadline, not an outage today, and a model-name swap alone won't preserve every agent's tool calls. OpenAI's deprecation notice also gives fine-tuned versions the same shutdown date.
Key Takeaways:
- Five dated GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini snapshots stop serving through the API on January 14, 2027.
- An older GPT-4o snapshot and GPT-4.1 nano have an earlier October 23, 2026 cutoff and different recommended replacements.
- GPT-6 Luna defaults to medium reasoning effort; Chat Completions function calling requires
reasoning_effortset tonone. - Existing eligible fine-tuning customers lose the ability to create new jobs on January 6, eight days before the January shutdown.
What OpenAI announced on October 14, 2026
OpenAI lists four families but five dated snapshots in its October 14 block. Deprecation means notice, not an outage or automatic redirection; an October success says nothing about January readiness.
Check the official table again before deploying.
Keep the work, not a fixed model name
Zentor routes tasks by fit and keeps workspace memory portable. Describe the recurring job you want done, then review early access.
Get early access →
Which model IDs shut down, and when?
| API model ID | Shutdown | OpenAI's recommended replacement |
|---|---|---|
gpt-4o-2024-08-06 |
Jan 14, 2027 | gpt-6-luna |
gpt-4o-2024-11-20 |
Jan 14, 2027 | gpt-6-luna |
gpt-4o-mini-2024-07-18 |
Jan 14, 2027 | gpt-6-luna |
gpt-4.1-2025-04-14 |
Jan 14, 2027 | gpt-6-luna |
gpt-4.1-mini-2025-04-14 |
Jan 14, 2027 | gpt-6-luna |
gpt-4o-2024-05-13 |
Oct 23, 2026 | gpt-5.6-sol |
gpt-4.1-nano (including gpt-4.1-nano-2025-04-14) |
Oct 23, 2026 | gpt-5.6-luna |
The last two rows come from the April 22 legacy notice, not the January announcement. Our October 23 migration guide covers that earlier batch. Keep two dates on your migration board. A job using the May 2024 GPT-4o snapshot needs attention this month, while a job pinned to the August or November 2024 snapshots has until January. For gpt-4.1-nano, the older notice explicitly lists the bare alias alongside its dated snapshot. The new notice names GPT-4.1 and GPT-4.1 mini, not nano.
Why GPT-6 Luna isn't a drop-in replacement
The Luna model card defaults to medium reasoning. Chat Completions function calling needs reasoning_effort: none; OpenAI recommends Responses for tools. Test the order-lookup call itself: a text-only demo can pass while tool execution fails.
For this gpt-4o api deprecation, test Chat Completions with reasoning_effort: none or move tool use to Responses. Neither branch guarantees equivalent latency or output. Zentor's Smart Routing is a product-design contrast, not a claim that Zentor supports Luna or migrates outside integrations.
What breaks in agents and saved automations
Find the ID each run sends in saved Zapier, n8n or Make steps and configuration. The GPT-4o and GPT-4.1 cards identify default snapshots gpt-4o-2024-08-06 and gpt-4.1-2025-04-14. Check live alias resolution.
Suppose you change a chat model but leave a scheduled invoice-summary step untouched. The demo works; the saved step still sends the retiring ID. Record its owner, request model, endpoint, tools and expected output. Take a sample of successful past runs and compare the successor's invoices: did it identify the same amounts, refuse the same ambiguous cases, call the lookup tool, and deliver to the same destination? A successful HTTP response alone doesn't answer those questions. Zentor's portable workspace-memory approach suggests keeping job context separate from model selection, an architectural lesson rather than a migration feature claim.
Fine-tuning has an earlier January deadline
Fine-tuned versions stop with their bases January 14. OpenAI's fine-tuning schedule ends new jobs for active existing customers January 6, 2027; earlier eligibility restrictions still apply. The Luna page says fine-tuning is not supported. Trained behavior won't transfer with the model name.
Preserve authorized training examples and outputs; test an untuned successor on the same cases. Luna's model card doesn't offer fine-tuning.
Costs and behavior: compare requests, not sticker prices
As of October 2026, OpenAI's model cards list standard text-token prices per million tokens below. These are provider API prices, not Zentor subscription prices.
| Model | Input / 1M | Output / 1M | Context window |
|---|---|---|---|
| GPT-4o | $2.50 | $10.00 | 128,000 |
| GPT-4.1 | $2.00 | $8.00 | 1,047,576 |
| GPT-4o mini | $0.15 | $0.60 | 128,000 |
| GPT-6 Luna | $0.10 | $0.50 | 1,050,000 |
Luna saves less against GPT-4o mini than GPT-4o. Prompts above 272,000 input tokens have higher rates; its maximum input is 922,000 tokens even though total context is 1,050,000. The knowledge cutoffs are October 2023 for GPT-4o, June 2024 for GPT-4.1 and May 2026 for Luna. The price table excludes caching, regional processing and special modes; it isn't a per-workflow invoice. If the new model makes more tool calls or generates longer answers, the cost per completed task can differ from a straight token-rate comparison. Measure actual billed usage and task outcomes, including reasoning; Zentor's Smart Routing also judges fit by task, not token price alone.
Migration checklist before January 14, 2027
- List live aliases, dated IDs, saved no-code steps and fine-tunes; assign each an owner.
- Fix October 23 IDs first; stage Luna for the January five.
- Replay tool calls, structured outputs and failures. Test Chat Completions with
reasoning_effort: noneor Responses with the required tools. - Compare quality, duration and billed usage; monitor request errors after rollout.
- Eligible fine-tuning users: finish permitted jobs before January 6 and migrate inference before January 14. Don't rely on a retired model for rollback.
Zentor's pricing page describes a workspace subscription, not a promise to migrate your OpenAI API integration or run a particular model.
FAQ: GPT-4o API retirement and migration
Does this affect ChatGPT?
The October 14 notice explicitly concerns API models. It isn't a new ChatGPT product-retirement announcement; check ChatGPT's own notices for its model availability.
Is GPT-4o fully gone?
No. The gpt-4o api deprecation names two dated snapshots for January 14, 2027, while gpt-4o-2024-05-13 has an October 23, 2026 deadline. Check your request ID rather than treating the entire family as one on/off switch.
What replaces GPT-4.1 nano?
OpenAI's older October 23 table recommends gpt-5.6-luna for gpt-4.1-nano; don't confuse that with gpt-6-luna, the recommended replacement for the January GPT-4.1 and GPT-4.1 mini snapshots.
How can I detect announced shutdowns programmatically?
OpenAI says to call GET /v1/models or retrieve an individual model and read the optional shutdown_date field. When present, it contains a YYYY-MM-DD date or null if no date was announced; a missing field is not proof of indefinite availability.
Zentor Editorial Zentor editorial team
The Zentor editorial team writes about workflow automation, AI agents, and the tools we build. Default byline for industry overviews, listicles, and collaborative pieces.
Ready to put this into practice?
Zentor runs browser tasks, research, and schedules automatically. Try it free.