diff --git a/agent/model_metadata.py b/agent/model_metadata.py
index 653a90619a5..956d6b93095 100644
--- a/agent/model_metadata.py
+++ b/agent/model_metadata.py
@@ -157,6 +157,13 @@ DEFAULT_CONTEXT_LENGTHS = {
     "gpt-5.4-nano": 400000,           # 400k (not 1.05M like full 5.4)
     "gpt-5.4-mini": 400000,           # 400k (not 1.05M like full 5.4)
     "gpt-5.4": 1050000,               # GPT-5.4, GPT-5.4 Pro (1.05M context)
+    # gpt-5.3-codex-spark is Codex-OAuth-only (ChatGPT Pro entitlement) and
+    # uses a smaller 128k window than other gpt-5.x slugs. Listed here as
+    # a defensive override so the longest-substring fallback doesn't match
+    # the generic "gpt-5" entry below (400k) and report the wrong limit if
+    # Spark's context ever needs to be resolved through this path. Real
+    # usage flows through _CODEX_OAUTH_CONTEXT_FALLBACK at line ~1113.
+    "gpt-5.3-codex-spark": 128000,
     "gpt-5.1-chat": 128000,           # Chat variant has 128k context
     "gpt-5": 400000,                  # GPT-5.x base, mini, codex variants (400k)
     "gpt-4.1": 1047576,
diff --git a/tests/hermes_cli/test_openai_codex_model_validation_fallback.py b/tests/hermes_cli/test_openai_codex_model_validation_fallback.py
index e33dbe2ba44..2b742b058ef 100644
--- a/tests/hermes_cli/test_openai_codex_model_validation_fallback.py
+++ b/tests/hermes_cli/test_openai_codex_model_validation_fallback.py
@@ -1,9 +1,18 @@
 """Regression tests for OpenAI Codex model validation when the listing lags behind
 actually usable backend model IDs.
 
-The bug: `/model` and `switch_model()` reject `gpt-5.3-codex-spark` because the
-OpenAI Codex listing omits it, even though direct runtime calls with
-`--provider openai-codex -m gpt-5.3-codex-spark` succeed.
+The bug originally reported in #16172: `/model` and `switch_model()` rejected
+`gpt-5.3-codex-spark` because the curated listing omitted it, even though direct
+runtime calls succeeded. PR #19729 fixed this by soft-accepting unknown-but-
+plausible Codex slugs with a warning, and this test pins the soft-accept
+behavior so it doesn't regress.
+
+Note: gpt-5.3-codex-spark itself is now in the curated catalog (PR #22991),
+so the real-world Spark request takes the `recognized=True` fast path. This
+test still uses Spark as the example slug but explicitly mocks
+``provider_model_ids`` to omit it, exercising the soft-accept path generically
+for any future entitlement-gated Codex slug that ships before Hermes catalogs
+it.
 """
 
 from unittest.mock import patch