diff --git a/opensquilla-webui/src/composables/setup/useSetupCatalog.privacy.test.ts b/opensquilla-webui/src/composables/setup/useSetupCatalog.privacy.test.ts
index d9f320e30f..2d7c021b79 100644
--- a/opensquilla-webui/src/composables/setup/useSetupCatalog.privacy.test.ts
+++ b/opensquilla-webui/src/composables/setup/useSetupCatalog.privacy.test.ts
@@ -6037,16 +6037,19 @@ describe('useSetupCatalog optional provider credentials', () => {
         source: 'not_required',
         probeReady: false,
       })
+      // The model id no longer gates the probe (#792): only the still-empty
+      // non-model required field (Base URL) blocks it.
       expect(credential?.probeDisabledReason).toBe(
-        'Complete required fields before verifying: Model, Base URL.',
+        'Complete required fields before verifying: Base URL.',
       )
 
       api.probeProviderConnection()
       expect(rpcCall.mock.calls.some(call => call[0] === 'onboarding.provider.probe')).toBe(false)
 
-      api.updateProviderField('model', 'test-model')
       api.updateProviderField('base_url', 'https://custom.example.test/v1')
       credential = api.providerPanel.value.credentialPanel
+      // An empty model is now allowed: reachability is verified via the
+      // model-list endpoint.
       expect(credential?.probeReady).toBe(true)
       expect(credential?.probeDisabledReason).toBe('')
 
@@ -6055,7 +6058,6 @@ describe('useSetupCatalog optional provider credentials', () => {
       expect(rpcCall).toHaveBeenCalledWith('onboarding.provider.probe', {
         providerId,
         baseUrl: 'https://custom.example.test/v1',
-        model: 'test-model',
       })
       app.unmount()
     },
diff --git a/opensquilla-webui/src/composables/setup/useSetupCatalog.ts b/opensquilla-webui/src/composables/setup/useSetupCatalog.ts
index d2bd3aa9ad..c81572bfb5 100644
--- a/opensquilla-webui/src/composables/setup/useSetupCatalog.ts
+++ b/opensquilla-webui/src/composables/setup/useSetupCatalog.ts
@@ -1548,19 +1548,20 @@ const providerProbeModel = computed(() => {
 
 const providerProbeMissingFields = computed(() => {
   if (!providerForm.selectedProvider.value) return []
+  // Stored/draft profiles probe through onboarding.llmProfile[.draft].probe,
+  // which still resolves a concrete deployment model, so keep requiring one.
   if (providerSelectionKind.value === 'profile') {
     return providerProbeModel.value ? [] : [t('setup.common.model')]
   }
+  // For a draft primary-provider config the model id no longer gates the
+  // probe: an empty model makes onboarding.provider.probe verify reachability
+  // via the model-list endpoint instead of a chat turn (#792).
   return providerFields.value
     .filter(field => field.required === true && !isProviderCredentialField(field))
-    .filter(field => {
-      const value = field.name === 'model'
-        && editingPrimaryProvider.value
-        && hasConfiguredPrimaryProvider.value
-        ? currentFormModelValue()
-        : providerForm.fieldValue(field, currentProviderConfig.value)
-      return !String(value ?? '').trim()
-    })
+    .filter(field => field.name !== 'model')
+    .filter(field => !String(
+      providerForm.fieldValue(field, currentProviderConfig.value) ?? '',
+    ).trim())
     .map(providerProbeFieldLabel)
 })
 
diff --git a/src/opensquilla/gateway/rpc_onboarding.py b/src/opensquilla/gateway/rpc_onboarding.py
index 1146fdffa7..5cc697ae51 100644
--- a/src/opensquilla/gateway/rpc_onboarding.py
+++ b/src/opensquilla/gateway/rpc_onboarding.py
@@ -57,9 +57,7 @@
 
 
 @contextmanager
-def _validation_error(
-    code: str, *, router_provider_id: str | None = None
-) -> Iterator[None]:
+def _validation_error(code: str, *, router_provider_id: str | None = None) -> Iterator[None]:
     """Translate a mutation validation error into a stable, client-localizable
     ``RpcHandlerError`` code, keeping the original English text as the message so
     the Web UI can fall back to it (and developers keep the detail).
@@ -115,6 +113,7 @@ def _channel_error() -> Iterator[None]:
             details={"fields": details} if details else None,
         ) from exc
 
+
 log = structlog.get_logger(__name__)
 
 _d = get_dispatcher()
@@ -166,9 +165,7 @@ def __init__(self, config: Any, usage_event_sink: Any) -> None:
         self._config = config
         self._usage_event_sink = usage_event_sink
 
-    async def probe_primary(
-        self, command: ProbePrimaryProvider
-    ) -> ProviderProbePayload:
+    async def probe_primary(self, command: ProbePrimaryProvider) -> ProviderProbePayload:
         return cast(
             "ProviderProbePayload",
             await _probe_primary_provider(
@@ -186,9 +183,7 @@ async def discover_primary_models(
             await _discover_primary_models(command, config=self._config),
         )
 
-    async def discover_image_models(
-        self, provider_id: str
-    ) -> ImageModelDiscoveryResult:
+    async def discover_image_models(self, provider_id: str) -> ImageModelDiscoveryResult:
         return cast(
             "ImageModelDiscoveryResult",
             await _discover_image_models(provider_id),
@@ -250,9 +245,7 @@ def __init__(
         self._connection_id = connection_id
         self._usage_event_sink = usage_event_sink
 
-    async def probe_saved(
-        self, command: ProfileProbeCommand
-    ) -> ProviderProbePayload:
+    async def probe_saved(self, command: ProfileProbeCommand) -> ProviderProbePayload:
         return cast(
             "ProviderProbePayload",
             await _probe_saved_profile(
@@ -263,9 +256,7 @@ async def probe_saved(
             ),
         )
 
-    async def probe_draft(
-        self, command: ProfileProbeCommand
-    ) -> ProviderProbePayload:
+    async def probe_draft(self, command: ProfileProbeCommand) -> ProviderProbePayload:
         return cast(
             "ProviderProbePayload",
             await _probe_draft_profile(
@@ -276,9 +267,7 @@ async def probe_draft(
             ),
         )
 
-    async def discover_saved(
-        self, command: ProfileProbeCommand
-    ) -> ProviderModelDiscoveryResult:
+    async def discover_saved(self, command: ProfileProbeCommand) -> ProviderModelDiscoveryResult:
         return cast(
             "ProviderModelDiscoveryResult",
             await _discover_saved_profile_models(
@@ -288,9 +277,7 @@ async def discover_saved(
             ),
         )
 
-    async def discover_draft(
-        self, command: ProfileProbeCommand
-    ) -> ProviderModelDiscoveryResult:
+    async def discover_draft(self, command: ProfileProbeCommand) -> ProviderModelDiscoveryResult:
         return cast(
             "ProviderModelDiscoveryResult",
             await _discover_draft_profile_models(
@@ -345,15 +332,11 @@ async def _models_discover(params: Any, ctx: RpcContext) -> dict[str, Any]:
     )
 
 
-async def _image_generation_models_discover(
-    params: Any, ctx: RpcContext
-) -> dict[str, Any]:
+async def _image_generation_models_discover(params: Any, ctx: RpcContext) -> dict[str, Any]:
 
     return cast(
         dict[str, Any],
-        await _provider_setup(ctx).discover_image_models(
-            str(_require(params, "providerId"))
-        ),
+        await _provider_setup(ctx).discover_image_models(str(_require(params, "providerId"))),
     )
 
 
@@ -393,23 +376,17 @@ async def _llm_profile_draft_probe(params: Any, ctx: RpcContext) -> dict[str, An
     )
 
 
-async def _llm_profile_models_discover(
-    params: Any, ctx: RpcContext
-) -> dict[str, Any]:
+async def _llm_profile_models_discover(params: Any, ctx: RpcContext) -> dict[str, Any]:
     return cast(
         dict[str, Any],
         await _profile_lifecycle(ctx).discover_models(_profile_probe_command(params)),
     )
 
 
-async def _llm_profile_draft_models_discover(
-    params: Any, ctx: RpcContext
-) -> dict[str, Any]:
+async def _llm_profile_draft_models_discover(params: Any, ctx: RpcContext) -> dict[str, Any]:
     return cast(
         dict[str, Any],
-        await _profile_lifecycle(ctx).discover_draft_models(
-            _profile_probe_command(params)
-        ),
+        await _profile_lifecycle(ctx).discover_draft_models(_profile_probe_command(params)),
     )
 
 
@@ -484,9 +461,9 @@ def _request_changes_active_provider_connection(params: Any, cfg: Any) -> bool:
         canonical_tokenrhythm_base_url,
     )
 
-    requested_provider = str(
-        params.get("providerId") or getattr(llm, "provider", "") or ""
-    ).strip().lower()
+    requested_provider = (
+        str(params.get("providerId") or getattr(llm, "provider", "") or "").strip().lower()
+    )
 
     comparisons = (
         ("apiKey", "api_key"),
@@ -540,9 +517,7 @@ async def _provider_configure(params: Any, ctx: RpcContext) -> dict[str, Any]:
             proxy=str(_param(params, "proxy", "")),
             preset_id=str(_param(params, "presetId", "")),
             router_action=str(_param(params, "routerAction", "preserve")),
-            image_generation_intent=str(
-                _param(params, "imageGenerationIntent", "preserve")
-            ),
+            image_generation_intent=str(_param(params, "imageGenerationIntent", "preserve")),
         )
         result = await _provider_setup(ctx).configure_primary(command)
     return cast(dict[str, Any], result.to_payload())
@@ -595,9 +570,7 @@ async def _llm_profile_upsert_and_activate(params: Any, ctx: RpcContext) -> dict
     if not isinstance(params, dict) or any(value is None for value in params.values()):
         # Generated optional Python fields use None for omission. The new
         # wire Contract excludes explicit null so keep absence and clear distinct.
-        raise RpcHandlerError(
-            "INVALID_REQUEST", "Invalid save-and-activate profile parameters"
-        )
+        raise RpcHandlerError("INVALID_REQUEST", "Invalid save-and-activate profile parameters")
     try:
         p = validate_upsert_and_activate_params(params)
     except ValidationError as exc:
@@ -614,9 +587,7 @@ async def _llm_profile_upsert_and_activate(params: Any, ctx: RpcContext) -> dict
                 api_key=p.get("apiKey"),
                 api_key_env=p.get("apiKeyEnv"),
                 api_key_env_pool=p.get("apiKeyEnvPool"),
-                keep_current_secret=p.get(
-                    "keepCurrentSecret", p.get("preserveApiKey", False)
-                ),
+                keep_current_secret=p.get("keepCurrentSecret", p.get("preserveApiKey", False)),
                 base_url=p.get("baseUrl"),
                 proxy=p.get("proxy"),
                 router_action=p.get("routerAction", "preserve"),
@@ -684,9 +655,7 @@ async def _llm_profile_active_remove(params: Any, ctx: RpcContext) -> dict[str,
     replacement_provider_id = str(_require(params, "replacementProviderId"))
     replacement_model = str(_param(params, "replacementModel", "") or "")
     router_action = str(_param(params, "routerAction", "preserve"))
-    image_generation_intent = str(
-        _param(params, "imageGenerationIntent", "preserve")
-    )
+    image_generation_intent = str(_param(params, "imageGenerationIntent", "preserve"))
     try:
         result = await _profile_lifecycle(ctx).remove_active(
             RemoveActiveProfile(
@@ -699,12 +668,8 @@ async def _llm_profile_active_remove(params: Any, ctx: RpcContext) -> dict[str,
         )
     except LlmProfileActivationError as exc:
         code_by_reason = {
-            "primary_pool_unsupported": (
-                "onboarding.llmProfile.primary_pool_unsupported"
-            ),
-            "router_provider_conflict": (
-                "onboarding.llmProfile.router_provider_conflict"
-            ),
+            "primary_pool_unsupported": ("onboarding.llmProfile.primary_pool_unsupported"),
+            "router_provider_conflict": ("onboarding.llmProfile.router_provider_conflict"),
         }
         raise RpcHandlerError(
             code_by_reason.get(exc.reason, "onboarding.llmProfile.invalid"),
@@ -763,12 +728,8 @@ async def _llm_profile_activate(params: Any, ctx: RpcContext) -> dict[str, Any]:
         )
     except LlmProfileActivationError as exc:
         code_by_reason = {
-            "primary_pool_unsupported": (
-                "onboarding.llmProfile.primary_pool_unsupported"
-            ),
-            "router_provider_conflict": (
-                "onboarding.llmProfile.router_provider_conflict"
-            ),
+            "primary_pool_unsupported": ("onboarding.llmProfile.primary_pool_unsupported"),
+            "router_provider_conflict": ("onboarding.llmProfile.router_provider_conflict"),
         }
         code = code_by_reason.get(exc.reason, "onboarding.llmProfile.invalid")
         details = {
@@ -1104,7 +1065,7 @@ async def _probe_primary_provider(
     config: Any,
     usage_event_sink: Any,
 ) -> dict[str, Any]:
-    """Live one-token probe of a candidate provider config (nothing is saved)."""
+    """Live probe of a candidate provider config without saving it."""
     provider_id = command.provider_id
     cfg = config
     api_key = str(command.api_key or "")
@@ -1145,19 +1106,42 @@ async def _probe_primary_provider(
         if not proxy:
             proxy = str(getattr(cfg.llm, "proxy", "") or "")
     model = str(command.model or "")
+    allow_default_api_key_env = not same_provider or reuse_stored_credentials
     with _validation_error("onboarding.provider.invalid"):
-        result = await _usage_accounted_provider_probe(
-            usage_event_sink,
-            provider_id=str(provider_id),
-            model=model,
-            api_key=api_key,
-            api_key_env=api_key_env,
-            base_url=base_url,
-            proxy=proxy,
-            allow_default_api_key_env=(
-                not same_provider or reuse_stored_credentials
-            ),
-        )
+        if model.strip():
+            result = await _usage_accounted_provider_probe(
+                usage_event_sink,
+                provider_id=str(provider_id),
+                model=model,
+                api_key=api_key,
+                api_key_env=api_key_env,
+                base_url=base_url,
+                proxy=proxy,
+                allow_default_api_key_env=allow_default_api_key_env,
+            )
+        else:
+            # A model is unnecessary for an endpoint/credential connectivity
+            # check; model discovery exercises that path without a chat turn.
+            from opensquilla.onboarding.probe import (
+                ProviderProbeResult,
+                discover_provider_models,
+            )
+
+            listing = await discover_provider_models(
+                provider_id=str(provider_id),
+                api_key=api_key,
+                api_key_env=api_key_env,
+                base_url=base_url,
+                proxy=proxy,
+                allow_default_api_key_env=allow_default_api_key_env,
+            )
+            result = ProviderProbeResult(
+                ok=listing.ok,
+                provider_id=str(provider_id),
+                model="",
+                failure_kind=listing.failure_kind,
+                message=listing.detail,
+            )
     saved_model = str(getattr(cfg.llm, "model", "") or "").strip()
     if (
         same_provider
@@ -1277,13 +1261,9 @@ async def _discover_primary_models(
             api_key_env=api_key_env,
             base_url=base_url,
             proxy=proxy,
-            allow_default_api_key_env=(
-                not same_provider or reuse_stored_credentials
-            ),
+            allow_default_api_key_env=(not same_provider or reuse_stored_credentials),
             force_refresh=force_refresh,
-            persist_catalog=(
-                same_provider and reuse_stored_credentials and not request_overrides
-            ),
+            persist_catalog=(same_provider and reuse_stored_credentials and not request_overrides),
             catalog_config=cfg,
         )
     return result.to_payload()
@@ -1386,9 +1366,7 @@ async def _channel_probe(params: Any, ctx: RpcContext) -> dict[str, Any]:
         "probeKind": "local_validation",
         "restartRequired": True,
         "entry": redact_channel_entry(type_name, normalized),
-        "warnings": [
-            "Configuration is locally valid; no provider connection was attempted."
-        ],
+        "warnings": ["Configuration is locally valid; no provider connection was attempted."],
     }
 
 
@@ -1481,9 +1459,7 @@ async def _audio_configure(params: Any, ctx: RpcContext) -> dict[str, Any]:
 async def _capability_reset(params: Any, ctx: RpcContext) -> dict[str, Any]:
 
     with _validation_error("onboarding.capability.invalid"):
-        result = await _capability_setup(ctx).reset(
-            str(_require(params, "capabilityId"))
-        )
+        result = await _capability_setup(ctx).reset(str(_require(params, "capabilityId")))
     return cast(dict[str, Any], result.to_payload())
 
 
@@ -1622,9 +1598,7 @@ async def _channel_disable(params: Any, ctx: RpcContext) -> dict[str, Any]:
     "onboarding.llmProfile.probe": _llm_profile_probe,
     "onboarding.llmProfile.draft.probe": _llm_profile_draft_probe,
     "onboarding.llmProfile.models.discover": _llm_profile_models_discover,
-    "onboarding.llmProfile.draft.models.discover": (
-        _llm_profile_draft_models_discover
-    ),
+    "onboarding.llmProfile.draft.models.discover": (_llm_profile_draft_models_discover),
     "onboarding.router.configure": _router_configure,
     "onboarding.ensemble.configure": _ensemble_configure,
     "onboarding.search.configure": _search_configure,
diff --git a/tests/test_gateway/test_rpc_onboarding.py b/tests/test_gateway/test_rpc_onboarding.py
index 2a9fe060fd..573aacfde4 100644
--- a/tests/test_gateway/test_rpc_onboarding.py
+++ b/tests/test_gateway/test_rpc_onboarding.py
@@ -347,9 +347,7 @@ def sync_primary(self, provider_config):
             sync_calls.append(provider_config)
 
     ctx = _admin_ctx()
-    ctx.config = GatewayConfig(
-        llm={"provider": "dashscope", "model": "qwen3.7-plus"}
-    )
+    ctx.config = GatewayConfig(llm={"provider": "dashscope", "model": "qwen3.7-plus"})
     ctx.config.config_path = str(tmp_path / "c.toml")
     ctx.provider_selector = FakeSelector()
 
@@ -377,10 +375,7 @@ def sync_primary(self, provider_config):
     assert res.error is None, res.error
     assert ctx.config.squilla_router.default_tier == "c0"
     assert ctx.config.squilla_router.tiers["c0"]["provider"] == "volcengine"
-    assert (
-        ctx.config.squilla_router.tiers["c0"]["model"]
-        == "doubao-seed-1-6-251015"
-    )
+    assert ctx.config.squilla_router.tiers["c0"]["model"] == "doubao-seed-1-6-251015"
     assert ctx.config.llm.provider == "dashscope"
     assert ctx.config.llm.model == "qwen3.7-plus"
     assert len(sync_calls) == 1
@@ -391,10 +386,7 @@ def sync_primary(self, provider_config):
     assert persisted["llm"]["provider"] == "dashscope"
     assert persisted["llm"]["model"] == "qwen3.7-plus"
     assert persisted["squilla_router"]["default_tier"] == "c0"
-    assert (
-        persisted["squilla_router"]["tiers"]["c0"]["model"]
-        == "doubao-seed-1-6-251015"
-    )
+    assert persisted["squilla_router"]["tiers"]["c0"]["model"] == "doubao-seed-1-6-251015"
 
 
 @pytest.mark.asyncio
@@ -503,9 +495,7 @@ async def test_router_catalog_rpc(tmp_path, monkeypatch):
 
 
 @pytest.mark.asyncio
-async def test_ensemble_configure_partial_payload_updates_and_persists(
-    tmp_path, monkeypatch
-):
+async def test_ensemble_configure_partial_payload_updates_and_persists(tmp_path, monkeypatch):
     monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
     from opensquilla.gateway.config import GatewayConfig
 
@@ -542,9 +532,7 @@ async def test_ensemble_configure_partial_payload_updates_and_persists(
 
 
 @pytest.mark.asyncio
-async def test_ensemble_configure_accepts_full_camel_case_payload(
-    tmp_path, monkeypatch
-):
+async def test_ensemble_configure_accepts_full_camel_case_payload(tmp_path, monkeypatch):
     monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
     res = await get_dispatcher().dispatch(
         "r1",
@@ -577,9 +565,7 @@ async def test_ensemble_configure_accepts_full_camel_case_payload(
 
 
 @pytest.mark.asyncio
-async def test_ensemble_configure_rejects_out_of_range_proposer_retries(
-    tmp_path, monkeypatch
-):
+async def test_ensemble_configure_rejects_out_of_range_proposer_retries(tmp_path, monkeypatch):
     monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
     res = await get_dispatcher().dispatch(
         "r1",
@@ -753,10 +739,7 @@ async def test_image_generation_configure_redacts_api_key(tmp_path, monkeypatch)
 
     data = tomllib.loads(target.read_text())
     assert data["image_generation"]["enabled"] is True
-    assert (
-        data["image_generation"]["primary"]
-        == "openrouter/google/gemini-3.1-flash-image-preview"
-    )
+    assert data["image_generation"]["primary"] == "openrouter/google/gemini-3.1-flash-image-preview"
     assert data["image_generation"]["providers"]["openrouter"]["api_key"] == "sk-or"
 
 
@@ -1134,10 +1117,7 @@ async def test_image_generation_configure_can_disable_legacy_invalid_config(
     assert data["image_generation"]["enabled"] is False
     assert data["image_generation"]["primary"] == "openrouter/google//image"
     assert data["image_generation"]["fallbacks"] == ["openai/"]
-    assert (
-        data["image_generation"]["providers"]["openrouter"]["base_url"]
-        == "not-a-url"
-    )
+    assert data["image_generation"]["providers"]["openrouter"]["base_url"] == "not-a-url"
 
 
 @pytest.mark.asyncio
@@ -1173,9 +1153,7 @@ async def test_onboarding_status_marks_legacy_image_endpoint_mismatch_degraded(
     ctx = _read_ctx()
     ctx.config = GatewayConfig()
     ctx.config.image_generation.enabled = True
-    ctx.config.image_generation.primary = (
-        "openrouter/google/gemini-3.1-flash-image-preview"
-    )
+    ctx.config.image_generation.primary = "openrouter/google/gemini-3.1-flash-image-preview"
     openrouter_provider = ctx.config.image_generation.providers.openrouter
     openrouter_provider.api_key = "sk-synthetic-image"
     openrouter_provider.base_url = "https://api.openai.com/v1"
@@ -1418,9 +1396,7 @@ async def test_memory_embedding_configure_updates_ctx_config(tmp_path, monkeypat
 
 
 @pytest.mark.asyncio
-async def test_memory_embedding_configure_auto_can_store_remote_fallback(
-    tmp_path, monkeypatch
-):
+async def test_memory_embedding_configure_auto_can_store_remote_fallback(tmp_path, monkeypatch):
     monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
     from opensquilla.gateway.config import GatewayConfig
 
@@ -1536,9 +1512,7 @@ async def test_provider_configure_does_not_persist_runtime_api_key(tmp_path, mon
 
 
 @pytest.mark.asyncio
-async def test_provider_configure_persists_explicit_replacement_for_env_key(
-    tmp_path, monkeypatch
-):
+async def test_provider_configure_persists_explicit_replacement_for_env_key(tmp_path, monkeypatch):
     monkeypatch.setenv("OPENROUTER_API_KEY", "startup-key")
     from opensquilla.gateway.config import GatewayConfig
 
@@ -1923,9 +1897,7 @@ async def fake_discover(**kwargs):
 
 
 @pytest.mark.asyncio
-async def test_models_discover_unverified_provider_stays_empty_without_build(
-    tmp_path, monkeypatch
-):
+async def test_models_discover_unverified_provider_stays_empty_without_build(tmp_path, monkeypatch):
     monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
 
     def _unexpected_build(*_args, **_kwargs):
@@ -1951,6 +1923,58 @@ def _unexpected_build(*_args, **_kwargs):
     }
 
 
+@pytest.mark.asyncio
+async def test_provider_probe_without_model_verifies_via_model_list(tmp_path, monkeypatch):
+    """An empty model probes reachability through the model-list endpoint
+    instead of raising ``Model is required`` (#792)."""
+    monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
+    _stub_openai_transport(
+        monkeypatch,
+        httpx.Response(
+            200,
+            headers={"content-type": "application/json"},
+            content=b'{"data": [{"id": "gpt-x", "context_length": 32000}]}',
+        ),
+    )
+    res = await get_dispatcher().dispatch(
+        "r1",
+        "onboarding.provider.probe",
+        {"providerId": "openrouter", "apiKey": "sk-test"},
+        _admin_ctx(),
+    )
+    assert res.error is None, res.error
+    assert res.payload["ok"] is True
+    assert res.payload["model"] == ""
+    assert res.payload["failureKind"] == ""
+    # No chat round-trip happened; the chat-only timings stay at their
+    # never-reached-the-network sentinels.
+    assert res.payload["latencyMs"] == 0
+    assert res.payload["firstResponseMs"] is None
+
+
+@pytest.mark.asyncio
+async def test_provider_probe_without_model_reports_auth_failure(tmp_path, monkeypatch):
+    """A model-less probe surfaces a bad key through the same envelope."""
+    monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
+    _stub_openai_transport(
+        monkeypatch,
+        httpx.Response(
+            401,
+            headers={"content-type": "application/json"},
+            content=b'{"error": {"message": "Incorrect API key provided"}}',
+        ),
+    )
+    res = await get_dispatcher().dispatch(
+        "r1",
+        "onboarding.provider.probe",
+        {"providerId": "openrouter", "apiKey": "sk-bad"},
+        _admin_ctx(),
+    )
+    assert res.error is None, res.error
+    assert res.payload["ok"] is False
+    assert res.payload["failureKind"] == "auth_invalid"
+
+
 @pytest.mark.asyncio
 async def test_image_models_discover_requires_admin_scope(tmp_path, monkeypatch):
     monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
@@ -1967,9 +1991,7 @@ async def test_image_models_discover_requires_admin_scope(tmp_path, monkeypatch)
 
 
 @pytest.mark.asyncio
-async def test_image_models_discover_returns_image_specific_catalog(
-    tmp_path, monkeypatch
-):
+async def test_image_models_discover_returns_image_specific_catalog(tmp_path, monkeypatch):
     monkeypatch.setenv("OPENSQUILLA_GATEWAY_CONFIG_PATH", str(tmp_path / "c.toml"))
 
     async def _discover(provider_id: str):
@@ -1982,8 +2004,7 @@ async def _discover(provider_id: str):
         }
 
     monkeypatch.setattr(
-        "opensquilla.onboarding.image_generation_model_discovery."
-        "discover_image_generation_models",
+        "opensquilla.onboarding.image_generation_model_discovery.discover_image_generation_models",
         _discover,
     )