OpenAI retired 9 API models on October 23, 2026: what to replace them with

The retirement date was October 23, 2026

The retired models were gpt-3.5-turbo, gpt-4, gpt-4-turbo, gpt-4.1-nano, gpt-4o-2024-05-13, o1, o1-pro, o3-mini and o4-mini. On the official page, the row of every one of them except gpt-4o-2024-05-13 also listed dated snapshot or alias identifiers next to the model name; a search limited to the short model name could miss calls that use them.

3 replacements took over the calls of the 9 retired models

Calls to gpt-4, gpt-4-turbo, gpt-4o-2024-05-13, o1, o1-pro and o3-mini moved to gpt-5.6-sol. Calls to gpt-3.5-turbo and o4-mini moved to gpt-5.6-terra. Calls to gpt-4.1-nano moved to gpt-5.6-luna.

One row carried a setting. For o1-pro, the official page announced gpt-5.6-sol with reasoning.mode: pro as the replacement. Sending those calls to gpt-5.6-sol without that setting would not have matched what the page listed.

The Output bill fell for four models and rose for five

All prices below are OpenAI list prices in dollars per million tokens, verified on the public OpenAI pricing page and checked on October 24, 2026. The models whose Output price fell on migration: gpt-4, gpt-4-turbo, o1 and o1-pro. The models whose Output price rose: gpt-3.5-turbo, gpt-4.1-nano, gpt-4o-2024-05-13, o3-mini and o4-mini.

Input followed its own path. From $5.00 for gpt-4o-2024-05-13 to $4.00 for gpt-5.6-sol per million tokens, Input went down while Output went up.

Our data holds no cached Input price for gpt-3.5-turbo. Its replacement gpt-5.6-terra has a cached Input price of $0.20 per million tokens.

From $0.40 for gpt-4.1-nano to $1.20 for gpt-5.6-luna per million tokens, Output changed by +200%. Cached Input went from $0.025 for gpt-4.1-nano to $0.020 for gpt-5.6-luna per million tokens.

o3-mini and o4-mini carried the same Input and Output list prices but not the same replacement. Calls to o3-mini moved to gpt-5.6-sol, at $20.00 per million tokens for Output. Calls to o4-mini moved to gpt-5.6-terra, at $12.00 per million tokens for Output.

Direct API users could price the switch; reseller and app users could not

For calls made through a reseller, the reseller's own cut-off date and replacement prices could differ from OpenAI's: our figures cover OpenAI as the only seller, so they say nothing about a reseller's margin or schedule.

The switch ran in phases that could be undone

1. List every call

   Each retired model name, and the dated identifiers shown in the official rows, had to be searched for in code and configuration.

2. Test a sample

   A sample of real prompts went to the announced replacement, and the answers were compared with the retired model's answers. For o1-pro, reasoning.mode: pro was set on gpt-5.6-sol.

3. Double-send a slice

   A small share of traffic went to both the old and the new model, and quality and the bill were compared on identical requests.

4. Switch the traffic

   All calls moved to the replacement, with the previous configuration kept ready to restore.

5. Keep a way back

   A rollback was possible only while the retired model still answered; our data gives no fallback after October 23, 2026.

The double-send phase was paid at both rates. For gpt-4, that meant paying gpt-4 at $60.00 and gpt-5.6-sol at $20.00 per million Output tokens on the same slice of traffic.