diff --git a/docetl/operations/utils.py b/docetl/operations/utils.py
index f4ce3c48..163015ed 100644
--- a/docetl/operations/utils.py
+++ b/docetl/operations/utils.py
@@ -615,7 +615,7 @@ def call_llm_with_cache(
             len(props) == 1
             and list(props.values())[0].get("type") == "string"
             and scratchpad is None
-            and "ollama" in model
+            and ("ollama" in model or "azure/gpt-4o-mini" in model)
         ):
             use_tools = False
 
@@ -635,7 +635,7 @@ def call_llm_with_cache(
                     "type": "function",
                     "function": {
                         "name": "send_output",
-                        "description": "Send structured output back to the user",
+                        "description": "Send output back to the user",
                         "strict": True,
                         "parameters": parameters,
                         "additionalProperties": False,
@@ -858,7 +858,7 @@ def call_llm_with_gleaning(
                         "type": "function",
                         "function": {
                             "name": "send_output",
-                            "description": "Send structured output back to the user",
+                            "description": "Send output back to the user",
                             "strict": True,
                             "parameters": parameters,
                             "additionalProperties": False,
diff --git a/docetl/runner.py b/docetl/runner.py
index c81e13dd..6be5da4d 100644
--- a/docetl/runner.py
+++ b/docetl/runner.py
@@ -54,13 +54,16 @@ def __init__(self, config: Dict, max_threads: int = None):
         # Check if output path is correctly formatted as JSON
         output_path = self.config.get("pipeline", {}).get("output", {}).get("path")
         if output_path:
-            if not output_path.lower().endswith(".json"):
+            if not (
+                output_path.lower().endswith(".json")
+                or output_path.lower().endswith(".csv")
+            ):
                 raise ValueError(
-                    f"Output path '{output_path}' is not a JSON file. Please provide a path ending with '.json'."
+                    f"Output path '{output_path}' is not a JSON or CSV file. Please provide a path ending with '.json' or '.csv'."
                 )
         else:
             raise ValueError(
-                "No output path specified in the configuration. Please provide an output path ending with '.json' in the configuration."
+                "No output path specified in the configuration. Please provide an output path ending with '.json' or '.csv' in the configuration."
             )
 
         self.syntax_check()
@@ -77,6 +80,11 @@ def __init__(self, config: Dict, max_threads: int = None):
                 all_ops_until_and_including_current = [
                     op_map[prev_op] for prev_op in step["operations"][:idx]
                 ] + [op_map[op_name]]
+                # If there's no model in the op, add the default model
+                for op in all_ops_until_and_including_current:
+                    if "model" not in op:
+                        op["model"] = self.default_model
+
                 all_ops_str = json.dumps(all_ops_until_and_including_current)
                 self.step_op_hashes[step["name"]][op_name] = hashlib.sha256(
                     all_ops_str.encode()
@@ -207,8 +215,16 @@ def save_output(self, data: List[Dict]):
         self.console.rule("[cyan]Saving Output[/cyan]")
         output_config = self.config["pipeline"]["output"]
         if output_config["type"] == "file":
-            with open(output_config["path"], "w") as file:
-                json.dump(data, file, indent=2)
+            if output_config["path"].lower().endswith(".json"):
+                with open(output_config["path"], "w") as file:
+                    json.dump(data, file, indent=2)
+            else:  # CSV
+                import csv
+
+                with open(output_config["path"], "w", newline="") as file:
+                    writer = csv.DictWriter(file, fieldnames=data[0].keys())
+                    writer.writeheader()
+                    writer.writerows(data)
             self.console.print(
                 f"[green italic]💾 Output saved to {output_config['path']}[/green italic]"
             )
diff --git a/poetry.lock b/poetry.lock
index 8eedfc0d..a5ae584d 100644
--- a/poetry.lock
+++ b/poetry.lock
@@ -1534,13 +1534,13 @@ requests = ">=2.20"
 
 [[package]]
 name = "litellm"
-version = "1.48.10"
+version = "1.49.0"
 description = "Library to easily interface with LLM API providers"
 optional = false
 python-versions = "!=2.7.*,!=3.0.*,!=3.1.*,!=3.2.*,!=3.3.*,!=3.4.*,!=3.5.*,!=3.6.*,!=3.7.*,>=3.8"
 files = [
-    {file = "litellm-1.48.10-py3-none-any.whl", hash = "sha256:752efd59747a0895f4695d025c66f0b2258d80a61175f7cfa41dbe4894ef95e1"},
-    {file = "litellm-1.48.10.tar.gz", hash = "sha256:0a4ff75da78e66baeae0658ad8de498298310a5efda74c3d840ce2b013e8401d"},
+    {file = "litellm-1.49.0-py3-none-any.whl", hash = "sha256:53711018b730f8a4262c11461b702b771e46e0c974f9c0bcd5b384b027308dd5"},
+    {file = "litellm-1.49.0.tar.gz", hash = "sha256:f5ef51b571b14de318fccdd6728f4e705aad68250f9ed374c7fe6c4e95d6c008"},
 ]
 
 [package.dependencies]
@@ -1549,7 +1549,7 @@ click = "*"
 importlib-metadata = ">=6.8.0"
 jinja2 = ">=3.1.2,<4.0.0"
 jsonschema = ">=4.22.0,<5.0.0"
-openai = ">=1.45.0"
+openai = ">=1.51.0"
 pydantic = ">=2.0.0,<3.0.0"
 python-dotenv = ">=0.2.0"
 requests = ">=2.31.0,<3.0.0"
@@ -2330,13 +2330,13 @@ files = [
 
 [[package]]
 name = "openai"
-version = "1.50.2"
+version = "1.51.2"
 description = "The official Python library for the openai API"
 optional = false
 python-versions = ">=3.7.1"
 files = [
-    {file = "openai-1.50.2-py3-none-any.whl", hash = "sha256:822dd2051baa3393d0d5406990611975dd6f533020dc9375a34d4fe67e8b75f7"},
-    {file = "openai-1.50.2.tar.gz", hash = "sha256:3987ae027152fc8bea745d60b02c8f4c4a76e1b5c70e73565fa556db6f78c9e6"},
+    {file = "openai-1.51.2-py3-none-any.whl", hash = "sha256:5c5954711cba931423e471c37ff22ae0fd3892be9b083eee36459865fbbb83fa"},
+    {file = "openai-1.51.2.tar.gz", hash = "sha256:c6a51fac62a1ca9df85a522e462918f6bb6bc51a8897032217e453a0730123a6"},
 ]
 
 [package.dependencies]