Why OpenAI scheduled it for shutdown
A dated snapshot like -0613 exists so a model's behavior can be frozen while newer versions ship alongside it. OpenAI eventually retires the old snapshots rather than maintaining every one indefinitely. Once the shutdown date above passes, a request that names gpt-4-0613 stops resolving and your call fails — there is no automatic fall-through to a newer GPT-4.
The failure mode is what makes this worth catching early: nothing about your code changes, so the break shows up only as a runtime error (or a silently degraded feature) on the day the snapshot goes away.
What to move to
OpenAI lists the current flagship as the successor. For most workloads that pinned gpt-4-0613 years ago, moving to the current generation is a straight upgrade — larger context window, lower price per token, and stronger instruction-following — not a downgrade you're forced into.
The exact replacement OpenAI recommends, and its live status, is shown at the top of this page and tracked on the Model Deprecation Radar.
Migration notes
The biggest gotcha isn't the model swap — it's the API surface that shipped with this snapshot. gpt-4-0613 introduced the original functions and function_call request parameters, which OpenAI later superseded with tools and tool_choice. Code old enough to pin this model often still uses the old parameters; the newer models expect the tools shape, and responses come back as tool_calls rather than a single function_call.
Grep your codebase for the literal string gpt-4-0613 (and any config or environment variable that holds it) rather than a family name like gpt-4 — the dated snapshot is what actually stops resolving.