From f5ace8824cf2aacf59abc673a003a6d5e3a2972f Mon Sep 17 00:00:00 2001
From: Daniel Han <danielhanchen@gmail.com>
Date: Sat, 23 May 2026 18:33:27 +0000
Subject: [PATCH] fix(openai): drop legacy MMDD snapshot suffixes from picker

Codex P2 follow-up on PR #5684 (providers.py:104): the new OpenAI
denylist only stripped the modern `-YYYY-MM-DD` dated form, so
legacy compact snapshots like `gpt-3.5-turbo-0125`, `gpt-4-0613`,
`gpt-4-1106-preview`, and `gpt-4-0125-preview` slipped through. The
listing returns both the snapshot and the canonical id; the picker
should surface only the canonical so users do not accidentally
select a deprecated snapshot.

Extends the dated-suffix branch with `-\d{4}(?:-preview)?$` to
cover the compact `MMDD` form and the `-MMDD-preview` variant.
The new rule is anchored to the end of the id so canonical chat
ids whose minor version happens to be a year-like number
(`gpt-4.5`, `gpt-5.5`, `o3`) are unaffected.

Added `test_openai_legacy_compact_snapshot_suffixes_are_dropped`
to pin both sides (8 snapshot ids dropped, 7 canonical ids kept).
10 / 10 filter tests pass locally.
---
 studio/backend/core/inference/providers.py    |  8 +++-
 .../tests/test_provider_registry_filters.py   | 47 +++++++++++++++++++
 2 files changed, 54 insertions(+), 1 deletion(-)

diff --git a/studio/backend/core/inference/providers.py b/studio/backend/core/inference/providers.py
index 9ec1db3e6d..82111e9c90 100644
--- a/studio/backend/core/inference/providers.py
+++ b/studio/backend/core/inference/providers.py
@@ -100,8 +100,14 @@ PROVIDER_REGISTRY: dict[str, dict[str, Any]] = {
             # Fine-tunes carry the user's tenant in the id.
             r"|^ft:"
             # Dated snapshot suffixes hide behind the canonical id
-            # which is also in the listing.
+            # which is also in the listing. Covers both the modern
+            # `-YYYY-MM-DD` dated form and the legacy compact
+            # 4-digit `MMDD` form (e.g. `gpt-3.5-turbo-0125`,
+            # `gpt-4-0613`, `gpt-4-1106-preview`) so the picker only
+            # surfaces the canonical id even when the listing still
+            # returns the snapshot copy.
             r"|-\d{4}-\d{2}-\d{2}$"
+            r"|-\d{4}(?:-preview)?$"
         ),
     },
     "anthropic": {
diff --git a/studio/backend/tests/test_provider_registry_filters.py b/studio/backend/tests/test_provider_registry_filters.py
index e061fffac9..bd766261dd 100644
--- a/studio/backend/tests/test_provider_registry_filters.py
+++ b/studio/backend/tests/test_provider_registry_filters.py
@@ -207,6 +207,53 @@ def test_openai_legacy_instruct_completion_ids_are_dropped():
     assert dropped == [], dropped
 
 
+def test_openai_legacy_compact_snapshot_suffixes_are_dropped():
+    """Legacy `-MMDD` snapshot suffixes (gpt-3.5-turbo-0125,
+    gpt-4-0613, gpt-4-1106-preview, etc.) hide behind the canonical
+    id which the listing also returns; surface only the canonical so
+    users do not pick a deprecated snapshot by accident. The
+    `-\\d{4}(?:-preview)?$` rule must not catch canonical ids whose
+    minor version happens to be a year-like number (e.g. gpt-4.5,
+    o3) -- those are tested as KEEP below."""
+    dropped = _apply(
+        "openai",
+        [
+            "gpt-3.5-turbo-0125",
+            "gpt-3.5-turbo-0301",
+            "gpt-3.5-turbo-16k-0613",
+            "gpt-4-0613",
+            "gpt-4-0314",
+            "gpt-4-32k-0613",
+            "gpt-4-1106-preview",
+            "gpt-4-0125-preview",
+        ],
+    )
+    assert dropped == [], dropped
+
+    # Canonical chat ids that share a digit-heavy tail must survive.
+    kept = _apply(
+        "openai",
+        [
+            "gpt-3.5-turbo",
+            "gpt-4o",
+            "gpt-4.5",
+            "gpt-5.5",
+            "gpt-5.5-mini",
+            "gpt-5.5-pro",
+            "o3",
+        ],
+    )
+    assert set(kept) >= {
+        "gpt-3.5-turbo",
+        "gpt-4o",
+        "gpt-4.5",
+        "gpt-5.5",
+        "gpt-5.5-mini",
+        "gpt-5.5-pro",
+        "o3",
+    }, kept
+
+
 def test_openai_search_preview_is_kept_search_api_is_dropped():
     # Pin the search-vs-search-api distinction so a future regex tweak
     # doesn't silently regress to dropping chat-with-search models.