Merge branch 'kortix-ai:main' into sheets-agent

backend/pyproject.toml

@@ -12,7 +12,7 @@ classifiers = [
 requires-python = ">=3.11"
 dependencies = [
   "python-dotenv==1.0.1",
-  "litellm==1.72.2",
+  "litellm==1.75.2",
   "click==8.1.7",
   "questionary==2.0.1",
   "requests==2.32.3",
@@ -74,3 +74,8 @@ repository = "https://github.com/kortix-ai/suna"
 
 [tool.uv]
 package = false
+
+[dependency-groups]
+dev = [
+    "orjson>=3.11.1",
+]

backend/services/llm.py

@@ -21,7 +21,9 @@ from utils.logger import logger
 from utils.config import config
 
 # litellm.set_verbose=True
-litellm.modify_params=True
+# Let LiteLLM auto-adjust params and drop unsupported ones (e.g., GPT-5 temperature!=1)
+litellm.modify_params = True
+litellm.drop_params = True
 
 # Constants
 MAX_RETRIES = 2
@@ -144,7 +146,9 @@ def prepare_params(
             logger.debug(f"Skipping max_tokens for Claude 3.7 model: {model_name}")
             # Do not add any max_tokens parameter for Claude 3.7
         else:
-            param_name = "max_completion_tokens" if 'o1' in model_name else "max_tokens"
+            is_openai_o_series = 'o1' in model_name
+            is_openai_gpt5 = 'gpt-5' in model_name
+            param_name = "max_completion_tokens" if (is_openai_o_series or is_openai_gpt5) else "max_tokens"
             params[param_name] = max_tokens
 
     # Add tools if provided
@@ -199,6 +203,10 @@ def prepare_params(
     # Apply Anthropic prompt caching (minimal implementation)
     # Check model name *after* potential modifications (like adding bedrock/ prefix)
     effective_model_name = params.get("model", model_name) # Use model from params if set, else original
+
+    # OpenAI GPT-5: drop unsupported temperature param (only default 1 allowed)
+    if "gpt-5" in effective_model_name and "temperature" in params and params["temperature"] != 1:
+        params.pop("temperature", None)
     if "claude" in effective_model_name.lower() or "anthropic" in effective_model_name.lower():
         messages = params["messages"] # Direct reference, modification affects params
 

backend/utils/constants.py

@@ -84,6 +84,22 @@ MODELS = {
         },
         "tier_availability": ["paid"]
     },
+    "openai/gpt-5": {
+        "aliases": ["gpt-5"],
+        "pricing": {
+            "input_cost_per_million_tokens": 1.25,
+            "output_cost_per_million_tokens": 10.00
+        },
+        "tier_availability": ["paid"]
+    },
+    "openai/gpt-5-mini": {
+        "aliases": ["gpt-5-mini"],
+        "pricing": {
+            "input_cost_per_million_tokens": 0.25,
+            "output_cost_per_million_tokens": 2.00
+        },
+        "tier_availability": ["paid"]
+    },
     "openai/gpt-4.1-mini": {
         "aliases": ["gpt-4.1-mini"],
         "pricing": {

frontend/src/components/thread/chat-input/_use-model-selection.ts

@@ -78,7 +78,7 @@ export const MODELS = {
   },
   'gpt-4.1': { 
     tier: 'premium', 
-    priority: 96,
+    priority: 92,
     recommended: false,
     lowQuality: false
   },
@@ -94,6 +94,18 @@ export const MODELS = {
     recommended: false,
     lowQuality: false
   },
+  'gpt-5': { 
+    tier: 'premium', 
+    priority: 99,
+    recommended: false,
+    lowQuality: false
+  },
+  'gpt-5-mini': { 
+    tier: 'premium', 
+    priority: 88,
+    recommended: false,
+    lowQuality: false
+  },
   'gemini-2.5-flash:thinking': { 
     tier: 'premium', 
     priority: 84,