Skip to content

OpenRAG is a comprehensive, single package Retrieval-Augmented Generation platform built on Langflow, Docling, and Opensearch.

Python

fix: exclude both luna generations and their dated snapshots (#2444)

Rico Furtadoe036eccaff6a
diff --git a/docs/docs/reference/configuration.mdx b/docs/docs/reference/configuration.mdx
index 88049184e..e1bdc350c 100644
--- a/docs/docs/reference/configuration.mdx
+++ b/docs/docs/reference/configuration.mdx
@@ -149,6 +149,16 @@ matching is against the id as the picker shows it — `gpt-3.5-turbo`, not
name also covers Azure's regional listings (`eu/gpt-5.6-luna`,
`us/gpt-5.6-luna`); write `eu/*` to reach one region on its own.
+A pattern without a wildcard matches that id and nothing else, so it does **not**
+cover the provider's dated snapshots — `gpt-5.6-luna` leaves
+`gpt-5.6-luna-2026-07-09` in the picker. To suppress a model and every snapshot
+of it, end the pattern with `*`.
+
+Excluding an id hides it from the pickers and from the live
+`/models/{provider}` lists. It does not unroute it: an operator who types the
+id by hand still reaches the model. Exclusion is a curation tool, not an access
+control.
+
Exclusions are per provider row. Azure OpenAI serves the same models as
OpenAI, so a model you want gone from both needs the pattern on both rows.
diff --git a/src/config/model_providers.yaml b/src/config/model_providers.yaml
index 6d7e760d1..0e9db29eb 100644
--- a/src/config/model_providers.yaml
+++ b/src/config/model_providers.yaml
@@ -40,7 +40,19 @@ providers:
saas: true
exclude_models:
- gpt-3.5-* # a whole generation, superseded and near retirement
- - gpt-5.6-luna # refuses function tools on /v1/chat/completions
+ # Refuses function tools on /v1/chat/completions. Both generations, and
+ # the trailing `*` for the dated snapshots (`gpt-5.6-luna-2026-07-09`):
+ # an exact pattern matches the undated id only, so a snapshot released
+ # later reappears in the picker on its own.
+ #
+ # LiteLLM's table reports `supports_function_calling: true` for every
+ # luna row, including the 5.6 that does not — which is why the exclusion
+ # is hand-maintained here rather than read off the catalogue. Treat a new
+ # luna generation as defective until someone proves otherwise with a real
+ # tool call: an id excluded here can still be typed by hand, so hiding a
+ # working model costs less than offering a broken one.
+ - gpt-5.6-luna*
+ - gpt-6-luna*
- name: ollama
display_name: Ollama
@@ -168,10 +180,12 @@ providers:
on_prem: true
saas: true
# Azure serves the same OpenAI models, so it needs the same exclusions.
- # The plain name covers the regional ids (`eu/`, `us/`, `global/`) too.
+ # The plain name covers the regional ids (`eu/`, `us/`, `global/`) too;
+ # see the openai row above for why both luna generations are wildcarded.
exclude_models:
- gpt-3.5-*
- - gpt-5.6-luna
+ - gpt-5.6-luna*
+ - gpt-6-luna*
# Azure lets you name a deployment anything, so LiteLLM's table cannot know
# yours. Declare them here and they appear in the pickers for everyone;
# otherwise each user has to type the name into the model box by hand.
diff --git a/tests/unit/known_failures.txt b/tests/unit/known_failures.txt
index 3d3cc8964..5d4b6c522 100644
--- a/tests/unit/known_failures.txt
+++ b/tests/unit/known_failures.txt
@@ -22,7 +22,6 @@ tests/unit/config/test_model_providers_config.py::test_shipped_defaults_keep_oll
tests/unit/config/test_model_providers_config.py::test_both_azure_providers_ship_in_every_mode
tests/unit/services/test_model_catalog.py::test_catalog_publishes_azure_ai_where_the_config_enables_it
tests/unit/services/test_model_catalog.py::test_azure_openai_is_the_row_that_carries_the_gpt_41_family
-tests/unit/services/test_model_catalog.py::test_the_shipped_config_excludes_on_azure_as_well_as_openai
tests/unit/services/test_skip_os_security_setup_startup.py::test_setup_runs_when_flag_false
tests/unit/services/test_skip_os_security_setup_startup.py::test_setup_skipped_when_flag_true
tests/unit/test_agent_sources_fence_stripped.py::test_fence_untrusted_text_escapes_embedded_end_delimiter
diff --git a/tests/unit/services/test_model_catalog.py b/tests/unit/services/test_model_catalog.py
index 67f8ee302..6d7186446 100644
--- a/tests/unit/services/test_model_catalog.py
+++ b/tests/unit/services/test_model_catalog.py
@@ -495,6 +495,40 @@ def test_the_shipped_config_excludes_on_azure_as_well_as_openai(monkeypatch) ->
assert not [model_id for model_id in ids if "gpt-3.5" in model_id or "luna" in model_id]
+def test_an_exclusion_without_a_wildcard_leaves_the_dated_snapshots(monkeypatch) -> None:
+ """Why the shipped luna patterns end in `*`.
+
+ A pattern with no wildcard matches that id alone, so the provider's dated
+ snapshot of the same model stays in the picker — which is how
+ `gpt-5.6-luna-2026-07-09` survived an exclusion of `gpt-5.6-luna`. Pinned
+ because the fix is a single character in a YAML file and reads like a typo.
+ """
+ monkeypatch.setenv("OPENRAG_RUN_MODE", "oss")
+ exact = model_catalog._excluded("gpt-5.6-luna-2026-07-09", ("gpt-5.6-luna",))
+ wildcard = model_catalog._excluded("gpt-5.6-luna-2026-07-09", ("gpt-5.6-luna*",))
+
+ assert not exact
+ assert wildcard
+ # The wildcard keeps covering the regional forms the bare name covered.
+ assert model_catalog._excluded("us/gpt-5.6-luna", ("gpt-5.6-luna*",))
+
+
+def test_both_luna_generations_are_excluded_from_both_rows(monkeypatch) -> None:
+ """LiteLLM's table cannot be trusted for this, so the config carries it.
+
+ Every luna row reports `supports_function_calling: true`, including the 5.6
+ that refuses them in practice — which is why the exclusion is
+ hand-maintained. A new generation is treated as defective until proven
+ otherwise, since an excluded id can still be typed by hand but a broken one
+ offered in the picker fails at agent runtime.
+ """
+ monkeypatch.setenv("OPENRAG_RUN_MODE", "oss")
+ for provider in ("openai", "azure"):
+ patterns = model_catalog.exclusions_for(provider)
+ assert any(pattern.startswith("gpt-5.6-luna") for pattern in patterns), provider
+ assert any(pattern.startswith("gpt-6-luna") for pattern in patterns), provider
+
+
@pytest.mark.asyncio
async def test_live_model_failure_does_not_block_other_provider_refreshes(monkeypatch):
calls: list[str] = []