Migrating off GPT-4o

GPT-4o has no announced OpenAI shutdown date, but it's several generations behind the current GPT-5.6 line and is the single highest-search-volume "legacy model" query we see. This is the migration that most people land on this page looking for.

The swap

// All AI Ask API
{
-  "model": "gpt-4o",
+  "model": "gpt-5.6-terra",
   "messages": [...]
}

Same change applies if you call OpenAI directly with its own SDK — only the model identifier changes; request/response shape is otherwise unaffected for a same-provider swap.

What actually changes

Pricing from our live pricing registry; speed from our benchmark leaderboard where measured.

Gotchas

• GPT-4o is a non-reasoning model. GPT-5.6 Terra supports an optional reasoning mode — leaving it off gives you a comparable latency profile to GPT-4o; turning it on trades latency for materially better multi-step accuracy at higher token cost. Decide explicitly rather than inheriting whatever the default is.
• GPT-4o's max output is capped at 16,384 tokens. GPT-5.6 Terra allows much longer completions — if you built retry/continuation logic around hitting that 16K ceiling, you can likely simplify it away.
• Pricing structure changed generation to generation, not just the headline number — re-check your blended cost estimate at your actual input:output ratio rather than assuming the 3:1 blended figure applies to your workload.
• If your integration parses OpenAI SDK response objects directly (not just our unified API), double-check any GPT-4o-specific fields your code depends on — the SDK response shape for reasoning-capable models includes additional fields non-reasoning GPT-4o responses never had.

Alternatives beyond the obvious successor

FAQ