GPT-5, GPT-5-mini support

Improvements to model name resolution
Improved instructions for multi-step workflows when continuation is available
Improved instructions for chat tool
Improved preferred model resolution, moved code from registry -> each provider
Updated tests

providers/base.py

@@ -4,7 +4,10 @@ import logging
 from abc import ABC, abstractmethod
 from dataclasses import dataclass, field
 from enum import Enum
-from typing import Any, Optional
+from typing import TYPE_CHECKING, Any, Optional
+
+if TYPE_CHECKING:
+    from tools.models import ToolModelCategory
 
 logger = logging.getLogger(__name__)
 
@@ -118,10 +121,10 @@ def create_temperature_constraint(constraint_type: str) -> TemperatureConstraint
         return FixedTemperatureConstraint(1.0)
     elif constraint_type == "discrete":
         # For models with specific allowed values - using common OpenAI values as default
-        return DiscreteTemperatureConstraint([0.0, 0.3, 0.7, 1.0, 1.5, 2.0], 0.7)
+        return DiscreteTemperatureConstraint([0.0, 0.3, 0.7, 1.0, 1.5, 2.0], 0.3)
     else:
         # Default range constraint (for "range" or None)
-        return RangeTemperatureConstraint(0.0, 2.0, 0.7)
+        return RangeTemperatureConstraint(0.0, 2.0, 0.3)
 
 
 @dataclass
@@ -154,24 +157,11 @@ class ModelCapabilities:
     # Custom model flag (for models that only work with custom endpoints)
     is_custom: bool = False  # Whether this model requires custom API endpoints
 
-    # Temperature constraint object - preferred way to define temperature limits
+    # Temperature constraint object - defines temperature limits and behavior
     temperature_constraint: TemperatureConstraint = field(
-        default_factory=lambda: RangeTemperatureConstraint(0.0, 2.0, 0.7)
+        default_factory=lambda: RangeTemperatureConstraint(0.0, 2.0, 0.3)
     )
 
-    # Backward compatibility property for existing code
-    @property
-    def temperature_range(self) -> tuple[float, float]:
-        """Backward compatibility for existing code that uses temperature_range."""
-        if isinstance(self.temperature_constraint, RangeTemperatureConstraint):
-            return (self.temperature_constraint.min_temp, self.temperature_constraint.max_temp)
-        elif isinstance(self.temperature_constraint, FixedTemperatureConstraint):
-            return (self.temperature_constraint.value, self.temperature_constraint.value)
-        elif isinstance(self.temperature_constraint, DiscreteTemperatureConstraint):
-            values = self.temperature_constraint.allowed_values
-            return (min(values), max(values))
-        return (0.0, 2.0)  # Fallback
-
 
 @dataclass
 class ModelResponse:
@@ -268,18 +258,15 @@ class ModelProvider(ABC):
             if not capabilities.supports_temperature:
                 return None
 
-            # Get temperature range
-            min_temp, max_temp = capabilities.temperature_range
+            # Use temperature constraint to get corrected value
+            corrected_temp = capabilities.temperature_constraint.get_corrected_value(requested_temperature)
 
-            # Clamp to valid range
-            if requested_temperature < min_temp:
-                logger.debug(f"Clamping temperature from {requested_temperature} to {min_temp} for model {model_name}")
-                return min_temp
-            elif requested_temperature > max_temp:
-                logger.debug(f"Clamping temperature from {requested_temperature} to {max_temp} for model {model_name}")
-                return max_temp
-            else:
-                return requested_temperature
+            if corrected_temp != requested_temperature:
+                logger.debug(
+                    f"Adjusting temperature from {requested_temperature} to {corrected_temp} for model {model_name}"
+                )
+
+            return corrected_temp
 
         except Exception as e:
             logger.debug(f"Could not determine effective temperature for {model_name}: {e}")
@@ -294,10 +281,10 @@ class ModelProvider(ABC):
         """
         capabilities = self.get_capabilities(model_name)
 
-        # Validate temperature
-        min_temp, max_temp = capabilities.temperature_range
-        if not min_temp <= temperature <= max_temp:
-            raise ValueError(f"Temperature {temperature} out of range [{min_temp}, {max_temp}] for model {model_name}")
+        # Validate temperature using constraint
+        if not capabilities.temperature_constraint.validate(temperature):
+            constraint_desc = capabilities.temperature_constraint.get_description()
+            raise ValueError(f"Temperature {temperature} is invalid for model {model_name}. {constraint_desc}")
 
     @abstractmethod
     def supports_thinking_mode(self, model_name: str) -> bool:
@@ -441,3 +428,28 @@ class ModelProvider(ABC):
         """
         # Base implementation: no resources to clean up
         return
+
+    def get_preferred_model(self, category: "ToolModelCategory", allowed_models: list[str]) -> Optional[str]:
+        """Get the preferred model from this provider for a given category.
+
+        Args:
+            category: The tool category requiring a model
+            allowed_models: Pre-filtered list of model names that are allowed by restrictions
+
+        Returns:
+            Model name if this provider has a preference, None otherwise
+        """
+        # Default implementation - providers can override with specific logic
+        return None
+
+    def get_model_registry(self) -> Optional[dict[str, Any]]:
+        """Get the model registry for providers that maintain one.
+
+        This is a hook method for providers like CustomProvider that maintain
+        a dynamic model registry.
+
+        Returns:
+            Model registry dict or None if not applicable
+        """
+        # Default implementation - most providers don't have a registry
+        return None