OpenAI sets Jan 14 shutdown for GPT-4o and GPT-4.1 in the API
The 4o and 4.1 workhorses get a hard off-date; OpenAI routes you to $0.10 Luna. A bigger legacy cull hits Oct 23, and a new trick sharpens small agents.
Copy markdown
Five snapshots, one January deadline
OpenAI set Jan 14, 2027 as the shutdown for gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini-2024-07-18, gpt-4.1-2025-04-14 and gpt-4.1-mini-2025-04-14, announced Oct 14. Fine-tunes built on those bases die on the same date, so any pinned 4o or 4.1 call in production needs a migration plan now.
Luna is the off-ramp, at $0.10/$0.50
OpenAI points all five models to gpt-6-luna, priced $0.10 input and $0.50 output per million tokens on the Standard tier — cheaper than 4o on paper. It's a newer reasoning-era model, though, so re-test your prompts, output format and latency before you repoint live traffic.
A second, bigger cull lands Oct 23
Eight days out, a legacy batch also shuts off: gpt-3.5-turbo-0125, the original gpt-4 snapshots, o1, o3-mini, o4-mini and gpt-image-1, with gpt-5.6 Sol/Terra/Luna and gpt-image-2.5 as the swaps. If an old side project still calls any of these, it stops working on the 23rd.
ActionOPD distills decisions, not tokens
A new paper lifts Qwen3-4B and 8B tool-use by 3.3 to 6.9 points across four out-of-distribution benchmarks using action-level on-policy distillation — needing no teacher logits and no shared tokenizer. That makes it a cheap way to harden your own small, local agent against tools it has never seen.