On October 14, 2026, OpenAI told developers that GPT-4o, GPT-4o mini, GPT-4.1, and GPT-4.1 mini will be removed from the OpenAI API on January 14, 2027. OpenAI's recommended replacement for all four is gpt-6-luna, and fine-tuned versions of these models shut down on the same date (OpenAI deprecations, as of October 2026). If a Slack bot, a Microsoft Teams bot, or an internal tool still calls one of these models through the API, it has about three months to move.
We first published this page in May 2024, when GPT-4o launched and teams wanted it in Slack. This update covers what is being retired, what GPT-6 Luna costs and does differently, and a checklist for moving a team bot before the deadline.
Which GPT-4 models are being retired
| Model | Snapshot IDs | Removed from the API | OpenAI's recommended replacement |
| GPT-4o | gpt-4o-2024-08-06, gpt-4o-2024-11-20 | January 14, 2027 | gpt-6-luna |
| GPT-4o mini | gpt-4o-mini-2024-07-18 | January 14, 2027 | gpt-6-luna |
| GPT-4.1 | gpt-4.1-2025-04-14 | January 14, 2027 | gpt-6-luna |
| GPT-4.1 mini | gpt-4.1-mini-2025-04-14 | January 14, 2027 | gpt-6-luna |
Source: OpenAI API deprecations page, entry dated October 14, 2026. Many integrations call the short alias (gpt-4o or gpt-4o-mini) rather than a dated snapshot. OpenAI's GPT-4o model page lists gpt-4o-2024-08-06 as the default snapshot behind the gpt-4o alias, so treat the alias as affected too.
The notice is about the API. Older GPT-4 and GPT-4 Turbo snapshots are on separate, earlier schedules listed on the same page.
What GPT-6 Luna costs compared with GPT-4o
| Model | Input | Cached input | Output | Context window |
| GPT-4o | $2.50 | $1.25 | $10.00 | 128K |
| GPT-4.1 | $2.00 | $0.50 | $8.00 | About 1M |
| GPT-4o mini | $0.15 | $0.075 | $0.60 | 128K |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 | About 1.05M |
Standard-tier list prices per 1M tokens, as of October 2026, from OpenAI's model pages for GPT-6 Luna, GPT-4o, GPT-4o mini, and GPT-4.1. GPT-6 Luna charges 2x for input and cached input and 1.5x for output on the whole request when a prompt goes over 272K input tokens.
On list price alone, GPT-6 Luna is cheaper than each of the four retiring models. List price is not cost per task, though. Luna is a reasoning model, and reasoning tokens are billed as output, so measure the cost of a real conversation before and after the switch.
What changes when you move from GPT-4o to GPT-6 Luna
Per OpenAI's GPT-6 Luna model page, as of October 2026:
- Reasoning effort. GPT-6 Luna accepts a reasoning effort of none, low, medium, high, xhigh, or max, and the default is medium. A bot that answered quickly on GPT-4o may feel slower at the default, so start with a lower effort for chat replies and raise it only where answers need more thought.
- Tool calling on Chat Completions. OpenAI's GPT-6 Luna model page says (as of October 2026) function calling on Chat Completions works only with reasoning effort set to none. If your Slack bot calls tools through Chat Completions, either set effort to none or move to the Responses API.
- Built-in tools need the Responses API. Web search, file search, code interpreter, and MCP servers are available to Luna through the Responses API.
- No fine-tuning. OpenAI's GPT-6 Luna page lists fine-tuning as not supported, and fine-tuned GPT-4o and GPT-4.1 models shut down with their base models. A bot that depends on a fine-tuned GPT-4o mini needs its behavior moved into instructions, examples, or retrieval.
- Inputs and outputs. Luna takes text and image input and returns text. If a workflow sends audio to GPT-4o, check OpenAI's audio and realtime models separately.
- Knowledge cutoff. Luna's training data runs to May 18, 2026, much later than GPT-4o's, so answers about recent tools and events may change.
A migration checklist for Slack and Teams bots
- Find every place a retiring model is named: code, environment variables, config files, no-code automation tools, and Slack or Teams apps someone set up years ago. Search for gpt-4o, gpt-4o-mini, gpt-4.1, and gpt-4.1-mini, including dated snapshot IDs.
- Pick a replacement per job, not one model for everything. OpenAI describes GPT-6 Luna as its most efficient model for focused, high-volume tasks. Agents that do long multi-step research or hard reasoning may need a larger model.
- Set the reasoning effort deliberately and test on real questions from the channel the bot serves, including the ones it used to get wrong.
- Check how the bot calls tools. On Chat Completions, tool calls need reasoning effort none; on the Responses API they do not.
- Compare cost per conversation for a week of normal traffic, not list price.
- Finish before January 14, 2027. After that date, the models are removed from the API.
Avoid the next forced migration: follow a tier, not a snapshot
Model retirements are now routine. As of October 2026, OpenAI's deprecations page lists more removals for 2027, including GPT-5.1, GPT-5.3-Codex, and GPT-5.4 nano on April 1, 2027. A team bot that names an exact model ID has to be edited every time.
In Runbear, Claude Agent SDK agents choose GPT models by tier: GPT Astra, Sol, Terra, and Luna. Each tier row carries a Stable badge and follows the latest available version in that tier. As of October 2026, the GPT Luna row follows GPT-6 Luna (gpt-6-luna), the model OpenAI recommends in place of GPT-4o (agent settings docs). To keep an agent on one exact release, choose it under Specific version instead, keeping in mind that specific versions can become unavailable when the provider retires them.
The same agent can run Claude or Gemini instead of GPT, and it works in both Slack and Microsoft Teams. On Runbear-provided access, usage is billed in Runbear credits. If the agent uses your own OpenAI API key, OpenAI bills that key.
In Slack, the agent always answers @mentions and direct messages; replying without an @mention and keyword triggers are optional per-channel settings. See the Slack integration page for setup.
FAQ
When does GPT-4o stop working in the OpenAI API?
On January 14, 2027. OpenAI announced the removal on October 14, 2026 for gpt-4o-2024-08-06 and gpt-4o-2024-11-20, along with GPT-4o mini, GPT-4.1, and GPT-4.1 mini (OpenAI deprecations, as of October 2026).
What replaces GPT-4o mini?
OpenAI's deprecations page recommends gpt-6-luna as the replacement for GPT-4o, GPT-4o mini, GPT-4.1, and GPT-4.1 mini. As of October 2026 it lists at $0.10 per million input tokens and $0.50 per million output tokens, below GPT-4o mini's $0.15 and $0.60.
Can I keep using a fine-tuned GPT-4o mini?
Not after January 14, 2027. Per OpenAI's deprecations page (as of October 2026), fine-tuned versions shut down on the same schedule as their base models, and OpenAI's GPT-6 Luna page lists fine-tuning as not supported.
Is GPT-6 Luna a drop-in replacement for GPT-4o?
Not quite. It is a reasoning model with a default effort of medium, function calling on Chat Completions requires reasoning effort none, and built-in tools need the Responses API. Test latency, answer quality, and cost on your own traffic before switching production bots.
How do I use GPT models in Slack without pinning a model ID?
Use a Slack agent that selects models by tier. In Runbear, pick a GPT tier row with a Stable badge in the agent's Model Selection settings, and the agent follows the latest available version in that tier.
Runbear runs shared AI agents in Slack and Microsoft Teams that your team can build without code. See how teams use shared agents, compare Runbear and ChatGPT, or read our list of the best AI agents for Slack.
