fix: replace hard orjson imports with json in core SDK files

orjson is an optional dependency (only in proxy extras), so bare
`import orjson` in core SDK files breaks users who haven't installed it.
Replace with stdlib json.dumps/json.loads which is always available.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

litellm/llms/custom_httpx/llm_http_handler.py

@@ -15,7 +15,6 @@ from typing import (
 )
 
 import httpx  # type: ignore
-import orjson
 from openai.types.file_deleted import FileDeleted
 
 import litellm
@@ -179,7 +178,7 @@ class BaseLLMHTTPHandler:
                     data=(
                         signed_json_body
                         if signed_json_body is not None
-                        else orjson.dumps(data).decode()
+                        else json.dumps(data)
                     ),
                     timeout=timeout,
                     stream=stream,
@@ -239,7 +238,7 @@ class BaseLLMHTTPHandler:
                     data=(
                         signed_json_body
                         if signed_json_body is not None
-                        else orjson.dumps(data).decode()
+                        else json.dumps(data)
                     ),
                     timeout=timeout,
                     stream=stream,

litellm/llms/openai_like/chat/handler.py

@@ -4,7 +4,7 @@ OpenAI-like chat completion handler
 For handling OpenAI-like chat completions, like IBM WatsonX, etc.
 """
 
-import orjson
+import json
 from typing import Any, Callable, Optional, Union
 
 import httpx
@@ -139,7 +139,7 @@ class OpenAILikeChatHandler(OpenAILikeBase):
             client=client,
             api_base=api_base,
             headers=headers,
-            data=orjson.dumps(data),
+            data=json.dumps(data),
             model=model,
             messages=messages,
             logging_obj=logging_obj,
@@ -185,7 +185,7 @@ class OpenAILikeChatHandler(OpenAILikeBase):
 
         try:
             response = await client.post(
-                api_base, headers=headers, data=orjson.dumps(data), timeout=timeout
+                api_base, headers=headers, data=json.dumps(data), timeout=timeout
             )
             response.raise_for_status()
         except httpx.HTTPStatusError as e:
@@ -350,7 +350,7 @@ class OpenAILikeChatHandler(OpenAILikeBase):
                     ),
                     api_base=api_base,
                     headers=headers,
-                    data=orjson.dumps(data),
+                    data=json.dumps(data),
                     model=model,
                     messages=messages,
                     logging_obj=logging_obj,
@@ -370,7 +370,7 @@ class OpenAILikeChatHandler(OpenAILikeBase):
                     client = HTTPHandler(timeout=timeout)  # type: ignore
                 try:
                     response = client.post(
-                        url=api_base, headers=headers, data=orjson.dumps(data)
+                        url=api_base, headers=headers, data=json.dumps(data)
                     )
                     response.raise_for_status()
 

litellm/llms/openai_like/chat/transformation.py

@@ -2,10 +2,10 @@
 OpenAI-like chat completion transformation
 """
 
+import json
 from typing import TYPE_CHECKING, Any, List, Optional, Tuple, Union
 
 import httpx
-import orjson
 
 from litellm.secret_managers.main import get_secret_str
 from litellm.types.llms.openai import AllMessageValues, ChatCompletionAssistantMessage
@@ -95,7 +95,7 @@ class OpenAILikeChatConfig(OpenAIGPTConfig):
         custom_llm_provider: Optional[str],
         base_model: Optional[str],
     ) -> ModelResponse:
-        response_json = orjson.loads(response.content)
+        response_json = json.loads(response.content)
         logging_obj.post_call(
             input=messages,
             api_key="",