What’s actually changing
This is the one figure in this series confirmed directly against a primary source, not paraphrased from third-party coverage. OpenAI’s own deprecations page states its retirement policy in plain terms:
“Unless safety or compliance concerns require a faster timeline, we provide the following minimum notice periods before model retirement: Generally available models: At least 6 months.”
Under the “2026-04-22: Legacy GPT model snapshots” notice on that same page, gpt-4, gpt-4-0613, o1, o1-pro, and o4-mini are listed with a shutdown date of October 23, 2026, with gpt-5.6-sol and gpt-5.6-terra named as the direct replacements. That is roughly six months of notice from the 2026-04-22 posting date, consistent with OpenAI’s own stated minimum.
Structural Comparison Matrix
| Operational Aspect | Before (until 2026-10-23) | After (2026-10-23 onward) |
|---|---|---|
gpt-4 / gpt-4-0613 calls | Respond normally | Return errors |
o1 / o1-pro calls | Respond normally | Return errors |
o4-mini calls | Respond normally | Return errors |
| Recommended model | Any of the above | gpt-5.6-sol or gpt-5.6-terra |
Fix it: migrate before the cutover, not after
A hard cutover means there is no graceful fallback once October 23, 2026 passes: a call to a retired model ID fails, full stop. The practical fix is a model-string swap to gpt-5.6-sol or gpt-5.6-terra, but treat that as the start of the migration, not the whole thing.
gpt-5.6-sol and gpt-5.6-terra use the o200k_base encoding, not the cl100k_base encoding the retiring models used, which changes token counts for identical prompts. See GPT-5.6’s Context Window for what that means for chunking and cost estimation. GPT-5.6 also changed prompt-caching behavior, covered in GPT-5.6 Ends Free Prompt-Cache Writes, so a migration that only swaps the model string and skips re-checking caching configuration will get a real, avoidable cost surprise.
Check your own numbers before the deadline
Use the free AI Token Counter to see how your actual prompts tokenize under the newer encoding before you migrate, not after, then run those counts through the LLM Pricing Calculator to see what the migration actually costs against GPT-5.6’s current rates.
Confirmed version
Fetched directly from OpenAI’s own API deprecations page, notice dated 2026-04-22, shutdown date 2026-10-23. This is a primary source, not third-party reporting. Browse more coverage in the AI Productivity archive, or start from The 2026 LLM Token & Pricing Reset hub.
