Skip to content

Use Claude Code, Codex, VSCode, Pi, and OpenCode (and 6 other harnesses) for free (1.3B+ free tokens) from your terminal, app, IDE, or phone, and now from the browser with native browser sessions (multi-harness + multi-model) like OpenClaw (voice supported + ToS friendly)

Python

patch: Migrate retired NVIDIA Super selections to Ultra (#2135)

Ali Khokhar3d2ae2306752
diff --git a/README.md b/README.md
index 181a6ecc..3f77b4a7 100644
--- a/README.md
+++ b/README.md
@@ -95,7 +95,7 @@ terminal open.
1. Create an API key at [build.nvidia.com/settings/api-keys](https://build.nvidia.com/settings/api-keys).
2. Open the Admin UI URL from the server log.
3. Open NVIDIA NIM in **Providers**, click **Add key**, paste the key, and click **Save**.
-4. Leave `MODEL` on the default `nvidia_nim/nvidia/nemotron-3-super-120b-a12b`, or search the model dropdown and select another model.
+4. Leave `MODEL` on the default `nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b`, or search the model dropdown and select another model.
5. Click **Apply** to save your model selection.
To protect the local proxy with a bearer token, enable **Proxy Authentication**
@@ -196,7 +196,7 @@ For another API endpoint, open **Providers → Custom providers → Add provider
| Provider | Admin UI setting | Example `MODEL` |
| --- | --- | --- |
-| [NVIDIA NIM](https://build.nvidia.com/settings/api-keys) | `NVIDIA_NIM_API_KEY` | `nvidia_nim/nvidia/nemotron-3-super-120b-a12b` |
+| [NVIDIA NIM](https://build.nvidia.com/settings/api-keys) | `NVIDIA_NIM_API_KEY` | `nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b` |
| [OpenRouter](https://openrouter.ai/keys) | `OPENROUTER_API_KEY` | `open_router/openrouter/free` |
| [Groq](https://console.groq.com/keys) | `GROQ_API_KEY` | `groq/llama-3.3-70b-versatile` |
| [ClinePass](https://docs.cline.bot/getting-started/clinepass) | `CLINE_API_KEY` | `cline_pass/cline-pass/kimi-k3` |
diff --git a/e2e/test_admin_providers.py b/e2e/test_admin_providers.py
index de566816..5240b66b 100644
--- a/e2e/test_admin_providers.py
+++ b/e2e/test_admin_providers.py
@@ -24,7 +24,7 @@ def test_retired_provider_is_absent_and_default_setup_remains_available(
close_provider(page)
page.get_by_role("button", name="Model Config", exact=True).click()
expect(page.locator("#field-MODEL")).to_have_value(
- "nvidia_nim/nvidia/nemotron-3-super-120b-a12b"
+ "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b"
)
diff --git a/smoke/README.md b/smoke/README.md
index 570ab8ff..7f30083c 100644
--- a/smoke/README.md
+++ b/smoke/README.md
@@ -112,7 +112,7 @@ uv run pytest smoke/product -n 0 -s --tb=short
```powershell
$env:FCC_LIVE_SMOKE = "1"
$env:FCC_SMOKE_TARGETS = "nvidia_nim_cli"
-$env:FCC_SMOKE_NIM_MODELS = "nvidia/nemotron-3.5-lightning-30b-a3b,moonshotai/kimi-k3,minimaxai/minimax-m3,nvidia/nemotron-3-super-120b-a12b"
+$env:FCC_SMOKE_NIM_MODELS = "nvidia/nemotron-3.5-lightning-30b-a3b,moonshotai/kimi-k3,minimaxai/minimax-m3,nvidia/nemotron-3-ultra-550b-a55b"
uv run pytest smoke/product -n 0 -s --tb=short
```
diff --git a/smoke/lib/config.py b/smoke/lib/config.py
index 9310a34d..6137292e 100644
--- a/smoke/lib/config.py
+++ b/smoke/lib/config.py
@@ -50,7 +50,7 @@ TARGET_ALIASES = {
SECRET_KEY_PARTS = ("KEY", "TOKEN", "SECRET", "WEBHOOK", "AUTH")
PROVIDER_SMOKE_DEFAULT_MODELS: dict[str, str] = {
- "nvidia_nim": "nvidia_nim/nvidia/nemotron-3-super-120b-a12b",
+ "nvidia_nim": "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b",
"azure_openai": "azure_openai/gpt-5.1",
"open_router": "open_router/nvidia/nemotron-3-super-120b-a12b:free",
"mistral": "mistral/devstral-small-latest",
@@ -113,7 +113,7 @@ NVIDIA_NIM_CLI_DEFAULT_MODELS: tuple[str, ...] = (
"nvidia/nemotron-3.5-lightning-30b-a3b",
"moonshotai/kimi-k3",
"minimaxai/minimax-m3",
- "nvidia/nemotron-3-super-120b-a12b",
+ "nvidia/nemotron-3-ultra-550b-a55b",
)
OPENROUTER_FREE_CLI_DEFAULT_MODELS: tuple[str, ...] = (
diff --git a/src/free_claude_code/application/routing.py b/src/free_claude_code/application/routing.py
index efac57cf..3afd203f 100644
--- a/src/free_claude_code/application/routing.py
+++ b/src/free_claude_code/application/routing.py
@@ -12,6 +12,7 @@ from free_claude_code.config.model_refs import (
is_retired_model_ref,
parse_model_name,
parse_provider_type,
+ replace_retired_model_ref,
)
from free_claude_code.config.provider_catalog import PROVIDER_CATALOG
from free_claude_code.config.reasoning import ReasoningPreference
@@ -207,6 +208,13 @@ class ModelRouter:
parse_model_name(self._settings.model),
decoded.force_reasoning_off if decoded is not None else False,
)
+ replacement = replace_retired_model_ref(candidate)
+ if replacement != candidate:
+ return (
+ parse_provider_type(replacement),
+ parse_model_name(replacement),
+ decoded.force_reasoning_off if decoded is not None else False,
+ )
if decoded is not None:
if decoded.provider_id not in self._settings.provider_ids:
if decoded.provider_id.startswith("custom_") or model_name.partition(
diff --git a/src/free_claude_code/config/constants.py b/src/free_claude_code/config/constants.py
index 6a290368..3d66fb02 100644
--- a/src/free_claude_code/config/constants.py
+++ b/src/free_claude_code/config/constants.py
@@ -1,6 +1,6 @@
"""Shared defaults used by config models and provider adapters."""
-DEFAULT_MODEL = "nvidia_nim/nvidia/nemotron-3-super-120b-a12b"
+DEFAULT_MODEL = "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b"
# Client context allocation when the provider reports no model limit.
DEFAULT_MODEL_CONTEXT_TOKENS = 200_000
diff --git a/src/free_claude_code/config/env_migrations.py b/src/free_claude_code/config/env_migrations.py
index 1525fa5e..c67943bd 100644
--- a/src/free_claude_code/config/env_migrations.py
+++ b/src/free_claude_code/config/env_migrations.py
@@ -311,7 +311,7 @@ def _log_model_repairs(before: Mapping[str, str], after: Mapping[str, str]) -> N
changed = sorted(key for key in before if before[key] != after.get(key))
if changed:
logger.info(
- "Repaired retired provider selections in managed config: {}",
+ "Repaired retired model selections in managed config: {}",
", ".join(changed),
)
diff --git a/src/free_claude_code/config/model_refs.py b/src/free_claude_code/config/model_refs.py
index 7b706764..cb12260d 100644
--- a/src/free_claude_code/config/model_refs.py
+++ b/src/free_claude_code/config/model_refs.py
@@ -21,6 +21,13 @@ from .reasoning import ReasoningPreference
RETIRED_PROVIDER_IDS = frozenset({"github_models"})
+def replace_retired_model_ref(model_ref: str) -> str:
+ """Replace the exact retired hosted endpoint, retaining independent providers."""
+ if model_ref.strip() == "nvidia_nim/nvidia/nemotron-3-super-120b-a12b":
+ return "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b"
+ return model_ref
+
+
class ModelFallback(BaseModel):
"""One ordered fallback's model, reasoning preference, and key number."""
@@ -97,6 +104,8 @@ def normalize_retired_model_settings(
normalized[key] = DEFAULT_MODEL if key == "MODEL" else ""
else:
normalized.pop(key)
+ elif key in normalized:
+ normalized[key] = replace_retired_model_ref(normalized[key])
try:
fallbacks = decode_model_fallbacks(normalized.get("MODEL_FALLBACKS"))
except ValueError:
@@ -105,6 +114,18 @@ def normalize_retired_model_settings(
retained = tuple(
entry for entry in fallbacks if not is_retired_model_ref(entry.model)
)
+ original_pairs = [(entry.model, entry.api_key_number) for entry in retained]
+ if len(original_pairs) != len(set(original_pairs)):
+ return normalized # Settings must still reject pre-existing duplicates.
+ migrated: list[ModelFallback] = []
+ seen: set[tuple[str, int]] = set()
+ for entry in retained:
+ model = replace_retired_model_ref(entry.model)
+ pair = (model, entry.api_key_number)
+ if pair not in seen:
+ seen.add(pair)
+ migrated.append(entry.model_copy(update={"model": model}))
+ retained = tuple(migrated)
if retained:
normalized["MODEL_FALLBACKS"] = encode_model_fallbacks(retained)
elif preserve_empty_overrides:
diff --git a/src/free_claude_code/providers/nvidia_nim/models.py b/src/free_claude_code/providers/nvidia_nim/models.py
index 41accc4d..16101ac3 100644
--- a/src/free_claude_code/providers/nvidia_nim/models.py
+++ b/src/free_claude_code/providers/nvidia_nim/models.py
@@ -7,6 +7,8 @@ an exclusion when hosted chat is verified to work again. New IDs remain visible.
HOSTED_MODEL_EXCLUSIONS = frozenset(
{
+ # Hosted endpoint marked Deprecated by NVIDIA, checked 2026-10-21.
+ "nvidia/nemotron-3-super-120b-a12b",
# Model-specific 404 on the tested hosted routes, checked 2026-10-13.
"01-ai/yi-large",
"adept/fuyu-8b",
diff --git a/tests/api/test_admin.py b/tests/api/test_admin.py
index 4ea4237b..889d6c16 100644
--- a/tests/api/test_admin.py
+++ b/tests/api/test_admin.py
@@ -1,5 +1,6 @@
import json
import mimetypes
+import os
from dataclasses import replace
from pathlib import Path
from unittest.mock import AsyncMock, patch
@@ -116,6 +117,49 @@ def test_admin_retirement_preview_apply_and_runtime_agree(monkeypatch, tmp_path)
)
+def test_admin_super_migration_preview_apply_and_runtime_agree(monkeypatch, tmp_path):
+ _set_home(monkeypatch, tmp_path)
+ _clear_process_config(monkeypatch)
+ old = "nvidia_nim/nvidia/nemotron-3-super-120b-a12b"
+ new = "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b"
+ env_file = tmp_path / ".fcc" / ".env"
+ env_file.parent.mkdir(parents=True)
+ env_file.write_text(
+ "FCC_CONFIG_SCHEMA=2\nMODEL=groq/default\nMODEL_OPUS=groq/managed\nPRIVATE=hidden\n",
+ encoding="utf-8",
+ )
+ monkeypatch.setenv("MODEL_OPUS", old)
+ app = create_test_app()
+ client = _local_client(app)
+ submission = {
+ "values": {
+ "MODEL": old,
+ "MODEL_SONNET": old,
+ "MODEL_OPUS": "deepseek/cannot-unlock",
+ "MODEL_FALLBACKS": f"{old},{new},groq/backup",
+ "REASONING_SONNET": "off",
+ }
+ }
+ response = post_config(client, "/admin/api/config/apply", json=submission)
+ assert response.status_code == 200
+ body = response.json()
+ assert body["applied"]
+ assert old not in body["env_preview"]
+ assert "PRIVATE=********" in body["env_preview"]
+ values = dotenv_values_from_file(env_file)
+ assert values["MODEL"] == values["MODEL_SONNET"] == new
+ assert values["MODEL_OPUS"] == "groq/managed"
+ assert values["REASONING_SONNET"] == "off"
+ assert values["PRIVATE"] == "hidden"
+ effective = provider_manager_for_app(app).current_settings()
+ assert effective.model == effective.model_sonnet == effective.model_opus == new
+ assert effective.model_fallbacks == (
+ ModelFallback(model=new),
+ ModelFallback(model="groq/backup"),
+ )
+ assert os.environ["MODEL_OPUS"] == old
+
+
@pytest.fixture(autouse=True)
def _offline_credential_checks(monkeypatch):
"""These tests exercise config persistence; probe HTTP has its own suite."""
@@ -2176,7 +2220,7 @@ def test_admin_never_reads_arbitrary_current_directory_env(monkeypatch, tmp_path
config = _local_client(app).get("/admin/api/config").json()
model_field = next(field for field in config["fields"] if field["key"] == "MODEL")
- assert model_field["value"] == "nvidia_nim/nvidia/nemotron-3-super-120b-a12b"
+ assert model_field["value"] == "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b"
assert model_field["source"] == "default"
response = post_config(
diff --git a/tests/api/test_api.py b/tests/api/test_api.py
index 7ff3c71c..83e9e14a 100644
--- a/tests/api/test_api.py
+++ b/tests/api/test_api.py
@@ -21,8 +21,18 @@ from tests.providers.support import immediate_admission, make_provider_config
@pytest.mark.parametrize("stream", [False, True])
@pytest.mark.parametrize("prefix", ["", "anthropic/", "claude-3-freecc-no-thinking/"])
-def test_retired_model_ingress_calls_default_provider(client, stream, prefix):
- original = f"{prefix}github_models/vendor/opus"
+ "ref,upstream",
+ [
+ ("github_models/vendor/opus", DEFAULT_MODEL.split("/", 1)[1]),
+ (
+ "nvidia_nim/nvidia/nemotron-3-super-120b-a12b",
+ "nvidia/nemotron-3-ultra-550b-a55b",
+ ),
+ ],
+)
+def test_retired_model_ingress_calls_replacement(client, stream, prefix, ref, upstream):
+ original = f"{prefix}{ref}"
_stream_messages_calls.clear()
response = client.post(
"/v1/messages",
@@ -36,7 +46,7 @@ def test_retired_model_ingress_calls_default_provider(client, stream, prefix):
assert response.status_code == 200
assert len(_stream_messages_calls) == 1
args, kwargs = _stream_messages_calls[0]
- assert args[0].model == DEFAULT_MODEL.split("/", 1)[1]
+ assert args[0].model == upstream
assert kwargs["response_model"] == original
if prefix == "claude-3-freecc-no-thinking/":
assert kwargs["reasoning"] == ReasoningPolicy.off()
diff --git a/tests/api/test_openai_responses.py b/tests/api/test_openai_responses.py
index f6ced7b4..125f8091 100644
--- a/tests/api/test_openai_responses.py
+++ b/tests/api/test_openai_responses.py
@@ -130,13 +130,23 @@ def test_responses_probe_endpoints_return_204(
@pytest.mark.parametrize("prefix", ["", "anthropic/", "claude-3-freecc-no-thinking/"])
-def test_retired_response_id_uses_default_provider(responses_client, prefix):
+ "ref,upstream",
+ [
+ ("github_models/vendor/opus", DEFAULT_MODEL.split("/", 1)[1]),
+ (
+ "nvidia_nim/nvidia/nemotron-3-super-120b-a12b",
+ "nvidia/nemotron-3-ultra-550b-a55b",
+ ),
+ ],
+)
+def test_retired_response_id_uses_replacement(responses_client, prefix, ref, upstream):
client, provider = responses_client
- original = f"{prefix}github_models/vendor/opus"
+ original = f"{prefix}{ref}"
response = client.post("/v1/responses", json={"model": original, "input": "hello"})
assert response.status_code == 200
assert "response.completed" in response.text
- assert provider.requests[0].model == DEFAULT_MODEL.split("/", 1)[1]
+ assert provider.requests[0].model == upstream
assert provider.stream_kwargs[0]["response_model"] == original
if prefix == "claude-3-freecc-no-thinking/":
assert provider.stream_kwargs[0]["reasoning"] == ReasoningPolicy.off()
diff --git a/tests/api/test_routes_utilities.py b/tests/api/test_routes_utilities.py
index 37645e62..b3f650b7 100644
--- a/tests/api/test_routes_utilities.py
+++ b/tests/api/test_routes_utilities.py
@@ -60,7 +60,17 @@ def test_count_tokens_endpoint(client):
@pytest.mark.parametrize("prefix", ["", "anthropic/", "claude-3-freecc-no-thinking/"])
-def test_count_tokens_retired_model_uses_configured_default(prefix):
+ "ref,target",
+ [
+ ("github_models/vendor/opus", "groq/default"),
+ (
+ "nvidia_nim/nvidia/nemotron-3-super-120b-a12b",
+ "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b",
+ ),
+ ],
+)
+def test_count_tokens_retired_model_uses_replacement(prefix, ref, target):
test_app = create_test_app(
Settings(MODEL="groq/default", MODEL_OPUS="deepseek/opus")
)
@@ -71,7 +81,7 @@ def test_count_tokens_retired_model_uses_configured_default(prefix):
response = TestClient(test_app).post(
"/v1/messages/count_tokens",
json={
- "model": f"{prefix}github_models/vendor/opus",
+ "model": f"{prefix}{ref}",
"messages": [{"role": "user", "content": "hi"}],
},
)
@@ -82,8 +92,8 @@ def test_count_tokens_retired_model_uses_configured_default(prefix):
for call in trace.call_args_list
if call.kwargs["stage"] == "routing"
)
- assert routed["provider_model_ref"] == "groq/default"
- assert routed["gateway_model"] == f"{prefix}github_models/vendor/opus"
+ assert routed["provider_model_ref"] == target
+ assert routed["gateway_model"] == f"{prefix}{ref}"
def test_count_tokens_error_returns_500(client):
diff --git a/tests/application/test_routing.py b/tests/application/test_routing.py
index 50c7edcb..c7b7622d 100644
--- a/tests/application/test_routing.py
+++ b/tests/application/test_routing.py
@@ -13,10 +13,77 @@ from free_claude_code.core.anthropic.models import (
MessagesRequest,
TokenCountRequest,
)
+from free_claude_code.core.gateway_model_ids import desktop_model_id
from free_claude_code.core.openai_responses import OpenAIResponsesRequest
from free_claude_code.core.reasoning import ReasoningControl, ReasoningEffort
+ "encoding,off",
+ [
+ ("", False),
+ ("anthropic/", False),
+ ("claude-3-freecc-no-thinking/", True),
+ ("desktop", False),
+ ("desktop", True),
+ ],
+)
+def test_retired_super_direct_ids_route_to_ultra_not_configured_default(encoding, off):
+ old = "nvidia_nim/nvidia/nemotron-3-super-120b-a12b"
+ new = "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b"
+ original = (
+ desktop_model_id(old, no_thinking=off)
+ if encoding == "desktop"
+ else encoding + old
+ )
+ settings = Settings(
+ MODEL="groq/default",
+ REASONING_POLICY="high",
+ MODEL_FALLBACKS=(
+ ModelFallback(model=new),
+ ModelFallback(model="groq/backup"),
+ ),
+ )
+ router = ModelRouter(settings)
+ route = router.resolve(original)
+ assert route.original_model == original
+ assert route.primary.provider_model_ref == new
+ assert route.primary.api_key_number == 1
+ assert [entry.target.provider_model_ref for entry in route.fallbacks] == [
+ "groq/backup"
+ ]
+ assert route.reasoning_preference is (
+ ReasoningPreference.OFF if off else ReasoningPreference.HIGH
+ )
+ messages = MessagesRequest(
+ model=original, max_tokens=100, messages=[Message(role="user", content="hi")]
+ )
+ responses = OpenAIResponsesRequest(model=original, input="hi")
+ count = TokenCountRequest(
+ model=original, messages=[Message(role="user", content="hi")]
+ )
+ for routed in (
+ router.resolve_messages_request(messages),
+ router.resolve_responses_request(responses),
+ router.resolve_token_count_request(count),
+ ):
+ assert routed.request.model == "nvidia/nemotron-3-ultra-550b-a55b"
+ assert routed.resolved == route
+ assert messages.model == responses.model == count.model == original
+
+
+ "ref",
+ [
+ "open_router/nvidia/nemotron-3-super-120b-a12b",
+ "nvidia_nim/nvidia/nemotron-3-super-120b-a12b-successor",
+ ],
+)
+def test_super_replacement_does_not_redirect_other_endpoints(ref):
+ route = ModelRouter(Settings()).resolve("anthropic/" + ref)
+ assert route.primary.provider_model_ref == ref
+
+
@pytest.mark.parametrize(
"prefix,off",
[("", False), ("anthropic/", False), ("claude-3-freecc-no-thinking/", True)],
diff --git a/tests/config/test_config.py b/tests/config/test_config.py
index 91863024..61ca921b 100644
--- a/tests/config/test_config.py
+++ b/tests/config/test_config.py
@@ -1,5 +1,6 @@
"""Contracts for pure Settings and canonical source composition."""
+import json
from enum import Enum
from pathlib import Path
from unittest.mock import patch
@@ -24,6 +25,7 @@ from free_claude_code.config.loader import (
from free_claude_code.config.model_refs import (
ModelFallback,
configured_chat_model_refs,
+ normalize_retired_model_settings,
parse_model_name,
parse_provider_type,
)
@@ -36,6 +38,151 @@ from free_claude_code.config.paths import (
from free_claude_code.config.reasoning import ReasoningPreference
from free_claude_code.config.settings import Settings
+_SUPER_REF = "nvidia_nim/nvidia/nemotron-3-super-120b-a12b"
+_ULTRA_REF = "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b"
+
+
+def test_default_claude_aliases_use_hosted_ultra():
+ settings = compose_settings_snapshot({}, {}).settings
+ router = ModelRouter(settings)
+ for tier in ("fable", "opus", "sonnet", "haiku"):
+ assert (
+ router.resolve(f"claude-{tier}-test").primary.provider_model_ref
+ == _ULTRA_REF
+ )
+
+
[email protected]("source", [ConfigSource.MANAGED, ConfigSource.PROCESS])
+ "key", ["MODEL", "MODEL_FABLE", "MODEL_OPUS", "MODEL_SONNET", "MODEL_HAIKU"]
+)
+def test_super_migration_preserves_explicit_selection_and_source(source, key):
+ values = {
+ key: f" {_SUPER_REF} ",
+ "REASONING_POLICY": "high",
+ "REASONING_OPUS": "off",
+ }
+ managed = {"MODEL": "groq/default", key: "groq/managed"}
+ process = {}
+ if source is ConfigSource.MANAGED:
+ managed.update(values)
+ else:
+ process.update(values)
+ before = managed.copy(), process.copy()
+ snapshot = compose_settings_snapshot(managed, process)
+ assert getattr(snapshot.settings, key.lower()) == _ULTRA_REF
+ assert snapshot.sources[key.lower()] is source
+ assert snapshot.settings.reasoning_policy is ReasoningPreference.HIGH
+ assert snapshot.settings.reasoning_opus is ReasoningPreference.OFF
+ if key != "MODEL":
+ assert snapshot.settings.model == "groq/default"
+ assert (managed, process) == before
+
+
[email protected]("source", [ConfigSource.MANAGED, ConfigSource.PROCESS])
[email protected]("super_first", [True, False])
+def test_super_fallback_migration_preserves_order_reasoning_and_distinct_keys(
+ source, super_first
+):
+ first, second = (
+ (_SUPER_REF, _ULTRA_REF) if super_first else (_ULTRA_REF, _SUPER_REF)
+ )
+ values = {
+ "MODEL": _SUPER_REF,
+ "NVIDIA_NIM_API_KEY": '["key-one", "key-two"]',
+ "MODEL_FALLBACKS": json.dumps(
+ [
+ {"model": first, "reasoning": "low"},
+ {"model": second, "reasoning": "high"},
+ {"model": _SUPER_REF, "reasoning": "off", "api_key_number": 2},
+ {"model": "groq/backup", "reasoning": "medium"},
+ ]
+ ),
+ }
+ snapshot = compose_settings_snapshot(
+ values if source is ConfigSource.MANAGED else {},
+ values if source is ConfigSource.PROCESS else {},
+ )
+ assert snapshot.settings.model_fallbacks == (
+ ModelFallback(model=_ULTRA_REF, reasoning=ReasoningPreference.LOW),
+ ModelFallback(
+ model=_ULTRA_REF, reasoning=ReasoningPreference.OFF, api_key_number=2
+ ),
+ ModelFallback(model="groq/backup", reasoning=ReasoningPreference.MEDIUM),
+ )
+ assert snapshot.sources["model_fallbacks"] is source
+ route = ModelRouter(snapshot.settings).resolve("claude-opus-test")
+ assert [
+ (entry.target.provider_model_ref, entry.target.api_key_number)
+ for entry in route.fallbacks
+ ] == [
+ (_ULTRA_REF, 2),
+ ("groq/backup", 1),
+ ]
+
+
+ "refs",
+ [
+ [_SUPER_REF, _SUPER_REF, _ULTRA_REF],
+ [_SUPER_REF, _ULTRA_REF, _ULTRA_REF],
+ [_SUPER_REF, "groq/backup", "groq/backup"],
+ ],
+)
+def test_super_migration_does_not_hide_preexisting_fallback_duplicates(refs):
+ with pytest.raises(ValidationError, match="duplicate model/key pairs"):
+ compose_settings_snapshot({"MODEL_FALLBACKS": json.dumps(refs)}, {})
+
+
+ "fallbacks",
+ [
+ f"{_SUPER_REF},",
+ "[invalid",
+ '[{"model":"groq/a","api_key_number":0}]',
+ '[{"model":"groq/a","reasoning":"invalid"}]',
+ ],
+)
+def test_super_migration_keeps_invalid_settings_invalid(fallbacks):
+ with pytest.raises(ValidationError):
+ compose_settings_snapshot(
+ {"MODEL": _SUPER_REF, "MODEL_FALLBACKS": fallbacks}, {}
+ )
+
+
+def test_model_repairs_leave_unaffected_fallback_text_unchanged():
+ values = {
+ "MODEL_FALLBACKS": '[{"model":"groq/backup","reasoning":"inherit","api_key_number":1}]',
+ "UNKNOWN": "keep",
+ }
+ assert (
+ normalize_retired_model_settings(values, preserve_empty_overrides=False)
+ == values
+ )
+
+
+ "model",
+ [
+ "open_router/nvidia/nemotron-3-super-120b-a12b",
+ "custom_00000000000000000000000000000000/nvidia/nemotron-3-super-120b-a12b",
+ _SUPER_REF + "-successor",
+ ],
+)
+def test_super_migration_leaves_other_endpoints_unchanged(model):
+ values = {
+ "MODEL": model,
+ "MODEL_OPUS": model,
+ "MODEL_FALLBACKS": json.dumps(
+ [{"model": model, "reasoning": "inherit", "api_key_number": 1}],
+ separators=(",", ":"),
+ ),
+ }
+ assert (
+ normalize_retired_model_settings(values, preserve_empty_overrides=False)
+ == values
+ )
+
def test_builtin_effort_is_from_client_with_inherited_tier_efforts():
snapshot = compose_settings_snapshot({}, {})
diff --git a/tests/config/test_env_migrations.py b/tests/config/test_env_migrations.py
index 723492f2..cd5e7d08 100644
--- a/tests/config/test_env_migrations.py
+++ b/tests/config/test_env_migrations.py
@@ -19,6 +19,77 @@ from free_claude_code.config.loader import ManagedConfigStore
from free_claude_code.config.model_refs import ModelFallback, decode_model_fallbacks
+ "schema", ["", "FCC_CONFIG_SCHEMA=1\n", "FCC_CONFIG_SCHEMA=2\n"]
+)
+def test_super_migration_persists_once_preserving_other_settings(
+ monkeypatch, tmp_path, schema
+):
+ managed, _ = _paths(monkeypatch, tmp_path)
+ managed.parent.mkdir(parents=True)
+ old = "nvidia_nim/nvidia/nemotron-3-super-120b-a12b"
+ new = "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b"
+ managed.write_text(
+ schema + f"MODEL={old}\nMODEL_OPUS={old}\nMODEL_SONNET=groq/other\n"
+ f"MODEL_FALLBACKS={old},groq/backup\nREASONING_OPUS=off\n"
+ "NVIDIA_NIM_API_KEY=keep-secret\nUNKNOWN_PRIVATE=keep-unknown\n",
+ encoding="utf-8",
+ )
+ store = ManagedConfigStore()
+ store.initialize({})
+ values = dotenv_values_from_file(managed)
+ assert values["MODEL"] == values["MODEL_OPUS"] == new
+ assert values["MODEL_SONNET"] == "groq/other"
+ assert values["REASONING_OPUS"] == "off"
+ assert decode_model_fallbacks(values["MODEL_FALLBACKS"]) == (
+ ModelFallback(model=new),
+ ModelFallback(model="groq/backup"),
+ )
+ assert (
+ store.read({}).settings.nvidia_nim_api_keys[0].get_secret_value()
+ == "keep-secret"
+ )
+ assert values["UNKNOWN_PRIVATE"] == "keep-unknown"
+ before = managed.read_bytes(), managed.stat().st_mtime_ns
+ store.initialize({})
+ assert (managed.read_bytes(), managed.stat().st_mtime_ns) == before
+
+
+def test_super_migration_imports_legacy_selection_without_editing_source(
+ monkeypatch, tmp_path
+):
+ managed, legacy = _paths(monkeypatch, tmp_path)
+ legacy.parent.mkdir(parents=True)
+ original = "MODEL=nvidia_nim/nvidia/nemotron-3-super-120b-a12b\nNVIDIA_NIM_API_KEY=keep-secret\n"
+ legacy.write_text(original, encoding="utf-8")
+ consolidate_managed_config({})
+ assert (
+ dotenv_values_from_file(managed)["MODEL"]
+ == "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b"
+ )
+ assert legacy.read_text(encoding="utf-8") == original
+
+
+def test_super_migration_atomic_write_failure_preserves_managed_file(
+ monkeypatch, tmp_path
+):
+ managed, _ = _paths(monkeypatch, tmp_path)
+ managed.parent.mkdir(parents=True)
+ original = (
+ b"FCC_CONFIG_SCHEMA=2\nMODEL=nvidia_nim/nvidia/nemotron-3-super-120b-a12b\n"
+ )
+ managed.write_bytes(original)
+
+ def fail_replace(*args):
+ raise OSError("injected replace failure")
+
+ monkeypatch.setattr(env_migrations.os, "replace", fail_replace)
+ with pytest.raises(OSError, match="injected"):
+ consolidate_managed_config({})
+ assert managed.read_bytes() == original
+ assert list(managed.parent.glob("*.tmp")) == []
+
+
@pytest.mark.parametrize("schema", ["", "FCC_CONFIG_SCHEMA=1\n"])
def test_retired_models_repair_managed_config_once_preserving_unknowns(
monkeypatch, tmp_path, schema
diff --git a/tests/contracts/test_smoke_config.py b/tests/contracts/test_smoke_config.py
index 64cce607..68aab420 100644
--- a/tests/contracts/test_smoke_config.py
+++ b/tests/contracts/test_smoke_config.py
@@ -1231,7 +1231,7 @@ def test_nvidia_nim_cli_default_models_are_normalized() -> None:
"nvidia_nim/nvidia/nemotron-3.5-lightning-30b-a3b",
"nvidia_nim/moonshotai/kimi-k3",
"nvidia_nim/minimaxai/minimax-m3",
- "nvidia_nim/nvidia/nemotron-3-super-120b-a12b",
+ "nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b",
)
assert set(refs.values()) == {"nvidia_nim_cli_default"}
diff --git a/tests/providers/test_model_discovery.py b/tests/providers/test_model_discovery.py
index 0c471d4d..d0c756e7 100644
--- a/tests/providers/test_model_discovery.py
+++ b/tests/providers/test_model_discovery.py
@@ -6,8 +6,10 @@ from unittest.mock import AsyncMock, patch
import pytest
+from free_claude_code.application.model_catalog import read_model_catalog
from free_claude_code.application.model_metadata import ProviderModelInfo
from free_claude_code.config.api_keys import normalize_api_keys
+from free_claude_code.config.loader import compose_settings_snapshot
from free_claude_code.config.model_refs import ModelFallback
from free_claude_code.config.nim import NimSettings
from free_claude_code.config.provider_catalog import (
@@ -142,12 +144,13 @@ async def test_nim_exclusions_apply_only_to_hosted_nim_discovery(
"01-ai/yi-large", # Observed model-specific 404.
"meta/muse-glimmer-30b",
"nvidia/cosmos-reason2-8b",
+ "nvidia/nemotron-3-super-120b-a12b", # Deprecated hosted endpoint.
"nvidia/ising-calibration-1-35b-a3b", # Explicit retirement 410.
"nvidia/llama-nemotron-embed-vl-1b-v2", # Working embedding service.
"nvidia/ai-synthetic-video-detector",
)
retained = (
- "nvidia/nemotron-3-super-120b-a12b",
+ "nvidia/nemotron-3-ultra-550b-a55b",
"nvidia/future-chat-model",
"01-ai/yi-large-successor", # Exact IDs, not substring matching.
"deepseek-ai/deepseek-v4.1-flash", # Timeouts are not exclusions.
@@ -169,6 +172,34 @@ async def test_nim_exclusions_apply_only_to_hosted_nim_discovery(
mock_list.assert_awaited_once_with()
+async def test_super_migration_and_hosted_discovery_publish_ultra_catalog():
+ old = "nvidia/nemotron-3-super-120b-a12b"
+ new = "nvidia/nemotron-3-ultra-550b-a55b"
+ config = make_provider_config(api_key="test-key", base_url=NVIDIA_NIM_DEFAULT_BASE)
+ with patch("free_claude_code.providers.openai_chat.client.AsyncOpenAI"):
+ provider = NvidiaNimProvider(
+ config, nim_settings=NimSettings(), admission=immediate_admission()
+ )
+ with patch.object(
+ provider._client.models,
+ "list",
+ new_callable=AsyncMock,
+ return_value=SimpleNamespace(
+ data=[SimpleNamespace(id=old), SimpleNamespace(id=new)]
+ ),
+ ):
+ infos = await provider.list_model_infos()
+ settings = compose_settings_snapshot({"MODEL": "nvidia_nim/" + old}, {}).settings
+ manager = _manager(settings)
+ manager.cache_model_infos("nvidia_nim", infos)
+ catalog = read_model_catalog(manager)
+ assert catalog.default_model_id == "nvidia_nim/" + new
+ assert [model.provider_model_ref for model in catalog.models] == [
+ "nvidia_nim/" + new
+ ]
+
+
@pytest.mark.asyncio
async def test_nim_exclusions_preserve_retained_metadata() -> None:
config = make_provider_config(api_key="test-key", base_url=NVIDIA_NIM_DEFAULT_BASE)