fix: max tokens param (#5174)

* max tokens param

* fix unit test

* fix unit test

backend/onyx/llm/chat_llm.py

@@ -24,6 +24,7 @@ from langchain_core.messages import SystemMessageChunk
 from langchain_core.messages.tool import ToolCallChunk
 from langchain_core.messages.tool import ToolMessage
 from langchain_core.prompt_values import PromptValue
+from litellm.utils import get_supported_openai_params
 
 from onyx.configs.app_configs import LOG_DANSWER_MODEL_INTERACTIONS
 from onyx.configs.app_configs import MOCK_LLM_RESPONSE
@@ -52,6 +53,8 @@ litellm.telemetry = False
 _LLM_PROMPT_LONG_TERM_LOG_CATEGORY = "llm_prompt"
 VERTEX_CREDENTIALS_FILE_KWARG = "vertex_credentials"
 VERTEX_LOCATION_KWARG = "vertex_location"
+LEGACY_MAX_TOKENS_KWARG = "max_tokens"
+STANDARD_MAX_TOKENS_KWARG = "max_completion_tokens"
 
 
 class LLMTimeoutError(Exception):
@@ -313,14 +316,22 @@ class DefaultMultiLLM(LLM):
 
         self._model_kwargs = model_kwargs
 
-    def log_model_configs(self) -> None:
-        logger.debug(f"Config: {self.config}")
+        self._max_token_param = LEGACY_MAX_TOKENS_KWARG
+        try:
+            params = get_supported_openai_params(model_name, model_provider)
+            if STANDARD_MAX_TOKENS_KWARG in (params or []):
+                self._max_token_param = STANDARD_MAX_TOKENS_KWARG
+        except Exception as e:
+            logger.warning(f"Error getting supported openai params: {e}")
 
     def _safe_model_config(self) -> dict:
         dump = self.config.model_dump()
         dump["api_key"] = mask_string(dump.get("api_key", ""))
         return dump
 
+    def log_model_configs(self) -> None:
+        logger.debug(f"Config: {self._safe_model_config()}")
+
     def _record_call(self, prompt: LanguageModelInput) -> None:
         if self._long_term_logger:
             self._long_term_logger.record(
@@ -393,11 +404,14 @@ class DefaultMultiLLM(LLM):
                 messages=processed_prompt,
                 tools=tools,
                 tool_choice=tool_choice if tools else None,
-                max_tokens=max_tokens,
                 # streaming choice
                 stream=stream,
                 # model params
-                temperature=self._temperature,
+                temperature=(
+                    1
+                    if self.config.model_name in ["gpt-5", "gpt-5-mini", "gpt-5-nano"]
+                    else self._temperature
+                ),
                 timeout=timeout_override or self._timeout,
                 # For now, we don't support parallel tool calls
                 # NOTE: we can't pass this in if tools are not specified
@@ -422,6 +436,7 @@ class DefaultMultiLLM(LLM):
                     if structured_response_format
                     else {}
                 ),
+                **({self._max_token_param: max_tokens} if max_tokens else {}),
                 **self._model_kwargs,
             )
         except Exception as e:

backend/onyx/llm/llm_provider_options.py

@@ -47,6 +47,9 @@ class WellKnownLLMProviderDescriptor(BaseModel):
 
 OPENAI_PROVIDER_NAME = "openai"
 OPEN_AI_MODEL_NAMES = [
+    "gpt-5",
+    "gpt-5-mini",
+    "gpt-5-nano",
     "o4-mini",
     "o3-mini",
     "o1-mini",
@@ -73,7 +76,14 @@ OPEN_AI_MODEL_NAMES = [
     "gpt-3.5-turbo-16k-0613",
     "gpt-3.5-turbo-0301",
 ]
-OPEN_AI_VISIBLE_MODEL_NAMES = ["o1", "o3-mini", "gpt-4o", "gpt-4o-mini"]
+OPEN_AI_VISIBLE_MODEL_NAMES = [
+    "gpt-5",
+    "gpt-5-mini",
+    "o1",
+    "o3-mini",
+    "gpt-4o",
+    "gpt-4o-mini",
+]
 
 BEDROCK_PROVIDER_NAME = "bedrock"
 # need to remove all the weird "bedrock/eu-central-1/anthropic.claude-v1" named

backend/onyx/onyxbot/slack/listener.py

@@ -130,6 +130,10 @@ _SLACK_GREETINGS_TO_IGNORE = {
 # This is always (currently) the user id of Slack's official slackbot
 _OFFICIAL_SLACKBOT_USER_ID = "USLACKBOT"
 
+# Fields to exclude from Slack payload logging
+# Intention is to not log slack message content
+_EXCLUDED_SLACK_PAYLOAD_FIELDS = {"text", "blocks"}
+
 
 class SlackbotHandler:
     def __init__(self) -> None:
@@ -570,6 +574,20 @@ class SlackbotHandler:
         sys.exit(0)
 
 
+def sanitize_slack_payload(payload: dict) -> dict:
+    """Remove message content from Slack payload for logging"""
+    sanitized = {
+        k: v for k, v in payload.items() if k not in _EXCLUDED_SLACK_PAYLOAD_FIELDS
+    }
+    if "event" in sanitized and isinstance(sanitized["event"], dict):
+        sanitized["event"] = {
+            k: v
+            for k, v in sanitized["event"].items()
+            if k not in _EXCLUDED_SLACK_PAYLOAD_FIELDS
+        }
+    return sanitized
+
+
 def prefilter_requests(req: SocketModeRequest, client: TenantSocketModeClient) -> bool:
     """True to keep going, False to ignore this Slack request"""
 
@@ -762,7 +780,10 @@ def prefilter_requests(req: SocketModeRequest, client: TenantSocketModeClient) -
     if not check_message_limit():
         return False
 
-    logger.debug(f"Handling Slack request: {client.bot_name=} '{req.payload=}'")
+    # Don't log Slack message content
+    logger.debug(
+        f"Handling Slack request: {client.bot_name=} '{sanitize_slack_payload(req.payload)=}'"
+    )
     return True
 
 
@@ -928,10 +949,9 @@ def process_message(
     if req.type == "events_api":
         event = cast(dict[str, Any], req.payload["event"])
         event_type = event.get("type")
-        msg = cast(str, event.get("text", ""))
         logger.info(
             f"process_message start: {tenant_id=} {req.type=} {req.envelope_id=} "
-            f"{event_type=} {msg=}"
+            f"{event_type=}"
         )
     else:
         logger.info(

backend/tests/unit/onyx/llm/test_chat_llm.py

@@ -148,7 +148,6 @@ def test_multiple_tool_calls(default_multi_llm: DefaultMultiLLM) -> None:
             ],
             tools=tools,
             tool_choice=None,
-            max_tokens=None,
             stream=False,
             temperature=0.0,  # Default value from GEN_AI_TEMPERATURE
             timeout=30,
@@ -294,7 +293,6 @@ def test_multiple_tool_calls_streaming(default_multi_llm: DefaultMultiLLM) -> No
             ],
             tools=tools,
             tool_choice=None,
-            max_tokens=None,
             stream=True,
             temperature=0.0,  # Default value from GEN_AI_TEMPERATURE
             timeout=30,

web/src/lib/hooks.ts

@@ -679,6 +679,8 @@ const MODEL_DISPLAY_NAMES: { [key: string]: string } = {
   "o1-mini": "o1 Mini",
   "o1-preview": "o1 Preview",
   o1: "o1",
+  "gpt-5": "GPT 5",
+  "gpt-5-mini": "GPT 5 Mini",
   "gpt-4.1": "GPT 4.1",
   "gpt-4": "GPT 4",
   "gpt-4o": "GPT 4o",