diff --git a/tests/llm_translation/test_fireworks_ai_translation.py b/tests/llm_translation/test_fireworks_ai_translation.py
index 1cc6aabdca..47b95c27ab 100644
--- a/tests/llm_translation/test_fireworks_ai_translation.py
+++ b/tests/llm_translation/test_fireworks_ai_translation.py
@@ -93,20 +93,32 @@ class TestFireworksAIAudioTranscription(BaseLLMAudioTranscriptionTest):
     [True, False],
 )
 def test_document_inlining_example(disable_add_transform_inline_image_block):
-    litellm.set_verbose = True
-    if disable_add_transform_inline_image_block is True:
-        with pytest.raises(Exception):
-            completion = litellm.completion(
-                model="fireworks_ai/accounts/fireworks/models/llama-v3p3-70b-instruct",
+    """
+    Document inlining appends ``#transform=inline`` to image/PDF URLs in the
+    outgoing request unless explicitly disabled. Assert the transform on the
+    serialized payload rather than making a live Fireworks call — the live
+    call only proved the model responded and broke whenever Fireworks rotated
+    its serverless model catalog.
+    """
+    from unittest.mock import patch
+
+    from litellm import completion
+    from litellm.llms.custom_httpx.http_handler import HTTPHandler
+
+    client = HTTPHandler()
+    pdf_url = "https://storage.googleapis.com/fireworks-public/test/sample_resume.pdf"
+
+    with patch.object(client, "post") as mock_post:
+        try:
+            completion(
+                model="fireworks_ai/accounts/fireworks/models/deepseek-v3p1",
                 messages=[
                     {
                         "role": "user",
                         "content": [
                             {
                                 "type": "image_url",
-                                "image_url": {
-                                    "url": "https://storage.googleapis.com/fireworks-public/test/sample_resume.pdf"
-                                },
+                                "image_url": {"url": pdf_url},
                             },
                             {
                                 "type": "text",
@@ -116,19 +128,19 @@ def test_document_inlining_example(disable_add_transform_inline_image_block):
                     }
                 ],
                 disable_add_transform_inline_image_block=disable_add_transform_inline_image_block,
+                client=client,
             )
-    else:
-        completion = litellm.completion(
-            model="fireworks_ai/accounts/fireworks/models/llama-v3p3-70b-instruct",
-            messages=[
-                {
-                    "role": "user",
-                    "content": "this is a test request, write a short poem",
-                },
-            ],
-            disable_add_transform_inline_image_block=disable_add_transform_inline_image_block,
-        )
-        print(completion)
+        except Exception as e:
+            print(e)
+
+        mock_post.assert_called_once()
+        json_data = json.loads(mock_post.call_args.kwargs["data"])
+        sent_url = json_data["messages"][0]["content"][0]["image_url"]["url"]
+        if disable_add_transform_inline_image_block is True:
+            assert sent_url == pdf_url
+            assert "#transform=inline" not in sent_url
+        else:
+            assert sent_url == pdf_url + "#transform=inline"
 
 
 @pytest.mark.parametrize(
@@ -215,7 +227,7 @@ def test_global_disable_flag_with_transform_messages_helper(monkeypatch):
     ) as mock_post:
         try:
             completion(
-                model="fireworks_ai/accounts/fireworks/models/llama-v3p3-70b-instruct",
+                model="fireworks_ai/accounts/fireworks/models/deepseek-v3p1",
                 messages=[
                     {
                         "role": "user",
diff --git a/tests/local_testing/test_completion.py b/tests/local_testing/test_completion.py
index 6341fa7800..cce6d33e79 100644
--- a/tests/local_testing/test_completion.py
+++ b/tests/local_testing/test_completion.py
@@ -1047,22 +1047,50 @@ def test_completion_openai_params(model):
 
 
 def test_completion_fireworks_ai():
-    try:
-        litellm.set_verbose = True
-        messages = [
-            {"role": "system", "content": "You're a good bot"},
+    """
+    Mocked so it does not depend on Fireworks' rotating serverless catalog
+    (no externally-verifiable model list exists). Asserts the request is
+    built correctly and the OpenAI-compatible response is parsed back.
+    """
+    litellm.set_verbose = True
+    messages = [
+        {"role": "system", "content": "You're a good bot"},
+        {"role": "user", "content": "Hey"},
+    ]
+
+    mock_response = MagicMock()
+    mock_response.status_code = 200
+    mock_response.headers = {"content-type": "application/json"}
+    mock_response.json.return_value = {
+        "id": "chatcmpl-test",
+        "object": "chat.completion",
+        "created": 1234567890,
+        "model": "accounts/fireworks/models/deepseek-v3p1",
+        "choices": [
             {
-                "role": "user",
-                "content": "Hey",
-            },
-        ]
+                "index": 0,
+                "message": {"role": "assistant", "content": "Hello there!"},
+                "finish_reason": "stop",
+            }
+        ],
+        "usage": {"prompt_tokens": 10, "completion_tokens": 2, "total_tokens": 12},
+    }
+    mock_response.text = json.dumps(mock_response.json.return_value)
+
+    client = HTTPHandler()
+    with patch.object(client, "post", return_value=mock_response) as mock_post:
         response = completion(
-            model="fireworks_ai/llama-v3p3-70b-instruct",
+            model="fireworks_ai/accounts/fireworks/models/deepseek-v3p1",
             messages=messages,
+            client=client,
         )
-        print(response)
-    except Exception as e:
-        pytest.fail(f"Error occurred: {e}")
+
+    mock_post.assert_called_once()
+    request_body = json.loads(mock_post.call_args.kwargs["data"])
+    assert "deepseek-v3p1" in request_body["model"]
+    assert request_body["messages"] == messages
+    assert response.choices[0].message.content == "Hello there!"
+    assert response.usage.total_tokens == 12
 
 
 @pytest.mark.parametrize(
diff --git a/tests/local_testing/test_completion_cost.py b/tests/local_testing/test_completion_cost.py
index 618287e195..cf0c645615 100644
--- a/tests/local_testing/test_completion_cost.py
+++ b/tests/local_testing/test_completion_cost.py
@@ -1171,7 +1171,7 @@ from litellm.llms.fireworks_ai.cost_calculator import get_base_model_for_pricing
 @pytest.mark.parametrize(
     "model, base_model",
     [
-        ("fireworks_ai/llama-v3p3-70b-instruct", "fireworks-ai-above-16b"),
+        ("fireworks_ai/llama-v3p1-70b-instruct", "fireworks-ai-above-16b"),
     ],
 )
 def test_get_model_params_fireworks_ai(model, base_model):
@@ -1182,18 +1182,47 @@ def test_get_model_params_fireworks_ai(model, base_model):
 @pytest.mark.parametrize(
     "model",
     [
-        "fireworks_ai/llama-v3p3-70b-instruct",
+        "fireworks_ai/accounts/fireworks/models/deepseek-v3p1",
     ],
 )
 def test_completion_cost_fireworks_ai(model):
+    """
+    Mocked so it does not depend on Fireworks' rotating serverless catalog.
+    Validates the Fireworks cost path: a parsed response with usage yields a
+    non-zero cost against the local cost map.
+    """
     os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
     litellm.model_cost = litellm.get_model_cost_map(url="")
 
-    messages = [{"role": "user", "content": "Hey, how's it going?"}]
-    resp = litellm.completion(model=model, messages=messages)  # works fine
+    mock_response_data = {
+        "id": "chatcmpl-test",
+        "object": "chat.completion",
+        "created": 1234567890,
+        "model": model.split("fireworks_ai/")[-1],
+        "choices": [
+            {
+                "index": 0,
+                "message": {"role": "assistant", "content": "Going great, thanks!"},
+                "finish_reason": "stop",
+            }
+        ],
+        "usage": {"prompt_tokens": 8, "completion_tokens": 5, "total_tokens": 13},
+    }
+
+    mock_response = MagicMock()
+    mock_response.status_code = 200
+    mock_response.headers = {"content-type": "application/json"}
+    mock_response.json.return_value = mock_response_data
+    mock_response.text = json.dumps(mock_response_data)
+
+    sync_handler = HTTPHandler()
+    messages = [{"role": "user", "content": "Hey, how's it going?"}]
+
+    with patch.object(HTTPHandler, "post", return_value=mock_response):
+        resp = litellm.completion(model=model, messages=messages, client=sync_handler)
 
-    print(resp)
     cost = completion_cost(completion_response=resp)
+    assert cost > 0
 
 
 def test_cost_azure_openai_prompt_caching():
diff --git a/tests/test_litellm/interactions/test_openapi_compliance.py b/tests/test_litellm/interactions/test_openapi_compliance.py
index 11d61d4c82..1d3b6b8ae1 100644
--- a/tests/test_litellm/interactions/test_openapi_compliance.py
+++ b/tests/test_litellm/interactions/test_openapi_compliance.py
@@ -153,12 +153,13 @@ class TestResponseCompliance:
 
     def test_interaction_response_fields(self, spec_dict):
         """Verify our InteractionsAPIResponse has correct fields."""
-        # The response is the Interaction schema
-        # Check CreateModelInteractionParams which includes output fields
-        schema = spec_dict["components"]["schemas"]["CreateModelInteractionParams"]
+        # The response is the dedicated `Interaction` schema. Google moved the
+        # output-only fields (notably the `steps` array, formerly `outputs`)
+        # off `CreateModelInteractionParams` and onto `Interaction`; the request
+        # schema no longer carries `steps`. Keep this aligned with the live spec.
+        schema = spec_dict["components"]["schemas"]["Interaction"]
 
-        # Output fields (readOnly). Google renamed `outputs` → `steps` in the
-        # upstream spec; keep this list aligned with the live schema.
+        # Output fields (readOnly).
         output_fields = [
             "id",
             "status",
@@ -175,7 +176,8 @@ class TestResponseCompliance:
 
     def test_status_enum_values(self, spec_dict):
         """Verify status enum values match spec."""
-        schema = spec_dict["components"]["schemas"]["CreateModelInteractionParams"]
+        # `status` is an output-only field; validate against the response schema.
+        schema = spec_dict["components"]["schemas"]["Interaction"]
         status_prop = schema["properties"]["status"]
         # Google Interactions API uses lowercase status values (updated Feb 2026)
         expected_statuses = [