diff --git a/apps/web/src/app/(onboarding)/setup/StepInferenceProvider.client.test.tsx b/apps/web/src/app/(onboarding)/setup/StepInferenceProvider.client.test.tsx
index 937606842..d606ee66a 100644
--- a/apps/web/src/app/(onboarding)/setup/StepInferenceProvider.client.test.tsx
+++ b/apps/web/src/app/(onboarding)/setup/StepInferenceProvider.client.test.tsx
@@ -249,6 +249,22 @@ function ollamaProviderStatus(): SetupModelStatus['providers'][number] {
   };
 }
 
+function openAiCompatibleProviderStatus(): SetupModelStatus['providers'][number] {
+  return {
+    id: 'openai-compatible',
+    label: 'OpenAI-compatible',
+    envVarName: 'OPENAI_COMPATIBLE_BASE_URL',
+    envVarLabel: 'Endpoint URL',
+    defaultRoomoteModel: '',
+    authKind: 'endpoint',
+    suggestedTaskModels: [],
+    additionalEnvFields: [],
+    additionalEnvValues: {},
+    runtimeApiKeySatisfied: false,
+    savedApiKeySatisfied: false,
+  };
+}
+
 function buildModelSetup(
   overrides: Partial<SetupModelStatus> = {},
 ): SetupModelStatus {
@@ -500,6 +516,54 @@ describe('StepInferenceProvider configured API key display', () => {
     expect(mutateAsyncMock).toHaveBeenCalledTimes(1);
   });
 
+  it('keeps setup incomplete when OpenAI-compatible discovery returns HTTP 400 with no models', async () => {
+    mutateAsyncMock.mockResolvedValueOnce({
+      error: 'OpenAI-compatible returned HTTP 400.',
+      modelCount: 0,
+      recommendedModels: [],
+    });
+    const onContinue = vi.fn();
+
+    render(
+      <StepInferenceProvider
+        modelSetup={buildModelSetup({
+          preselectedProvider: 'openai-compatible',
+          providers: [
+            openAiCompatibleProviderStatus(),
+            openrouterProviderStatus(),
+          ],
+        })}
+        onContinue={onContinue}
+      />,
+    );
+
+    selectProvider('openai-compatible');
+    fireEvent.change(
+      screen.getByPlaceholderText(/Endpoint URL for OpenAI-compatible/i),
+      { target: { value: 'https://proxy.example.com/v1' } },
+    );
+    fireEvent.change(
+      screen.getByRole('textbox', {
+        name: 'Connection name for OpenAI-compatible endpoint',
+      }),
+      { target: { value: 'company-proxy' } },
+    );
+    fireEvent.click(screen.getByRole('button', { name: /continue/i }));
+
+    await waitFor(() => {
+      expect(toast.error).toHaveBeenCalledWith(
+        'OpenAI-compatible returned HTTP 400.',
+      );
+    });
+    expect(mutateAsyncMock).toHaveBeenCalledOnce();
+    expect(mutateAsyncMock).toHaveBeenCalledWith({
+      provider: 'openai-compatible',
+      baseUrl: 'https://proxy.example.com/v1',
+      apiKey: undefined,
+    });
+    expect(onContinue).not.toHaveBeenCalled();
+  });
+
   it('reports a tool-calling failure separately from model eligibility', async () => {
     mutateAsyncMock
       .mockResolvedValueOnce({
diff --git a/apps/web/src/trpc/commands/task-models/local-provider-discovery.test.ts b/apps/web/src/trpc/commands/task-models/local-provider-discovery.test.ts
index 090b5d0a3..ba65ef897 100644
--- a/apps/web/src/trpc/commands/task-models/local-provider-discovery.test.ts
+++ b/apps/web/src/trpc/commands/task-models/local-provider-discovery.test.ts
@@ -166,6 +166,34 @@ describe('discoverProviderModels', () => {
     );
   });
 
+  it('returns no models when an OpenAI-compatible endpoint responds with HTTP 400', async () => {
+    fetchMock.mockResolvedValue(
+      new Response(JSON.stringify({ error: 'No models available' }), {
+        status: 400,
+        headers: { 'content-type': 'application/json' },
+      }),
+    );
+
+    await expect(
+      discoverProviderModels({
+        provider: 'openai-compatible',
+        baseUrl: 'https://proxy.example.com/v1',
+        apiKey: 'submitted-key',
+      }),
+    ).resolves.toEqual({
+      models: [],
+      modelCount: 0,
+      recommendedModels: [],
+      error: 'OpenAI-compatible returned HTTP 400.',
+    });
+    expect(fetchMock).toHaveBeenCalledWith(
+      'https://proxy.example.com/v1/models',
+      expect.objectContaining({
+        headers: { Authorization: 'Bearer submitted-key' },
+      }),
+    );
+  });
+
   it('uses saved LiteLLM credentials and metadata when discovering models', async () => {
     mockGetPersistedEnvironmentVariableValues.mockResolvedValue({
       LITELLM_BASE_URL: 'https://litellm.example/v1',