Overview
The model_config module provides configuration for different LLM models including temperature settings, display names, and model-specific parameters. It includes helper functions for accessing model configurations and selecting models for benchmarking.
Configuration Constants
MODEL_CONFIGS
Dictionary mapping BAMLModel enum values to model-specific configurations.
from model_config import MODEL_CONFIGS
from agents.llm import BAMLModel
# Access model configuration
config = MODEL_CONFIGS[BAMLModel.GPT4O_MINI]
print(config) # {'temperature': 0.7}
{
BAMLModel.GPT4O_MINI: {"temperature": 0.7},
BAMLModel.O3_MINI: {"temperature": 1.0}, # Reasoning models require 1.0
# ... more models
}
RESTRICTED_TEMPERATURE_MODELS
Set of models that require temperature 1.0 (OpenAI reasoning models).
from model_config import RESTRICTED_TEMPERATURE_MODELS
from agents.llm import BAMLModel
# Check if model has restricted temperature
if BAMLModel.O3_MINI in RESTRICTED_TEMPERATURE_MODELS:
print("This model requires temperature 1.0")
O4_MINIO3_MINIO3O1O1_MINIO1_PREVIEW
Functions
get_model_config()
Get configuration dictionary for a specific model.
def get_model_config(model: BAMLModel) -> Dict[str, Any]
Model enum value from agents.llm.BAMLModel
{"temperature": 0.7} if model not found.
get_temperature()
Get the appropriate temperature for a model.
def get_temperature(model: BAMLModel) -> float
is_temperature_restricted()
Check if a model has temperature restrictions.
def is_temperature_restricted(model: BAMLModel) -> bool
True if model requires temperature 1.0, False otherwise
get_benchmark_models()
Get the list of models for benchmarking.
def get_benchmark_models() -> list
BAMLModel enum values representing diverse models confirmed available as of December 2025.
Default benchmark models:
OPENROUTER_DEVSTRALOPENROUTER_MIMO_V2_FLASHOPENROUTER_NEMOTRON_NANOOPENROUTER_DEEPSEEK_R1T_CHIMERAOPENROUTER_DEEPSEEK_R1T2_CHIMERAOPENROUTER_GLM_45_AIROPENROUTER_LLAMA_33_70BOPENROUTER_OLMO3_32B
get_model_display_name()
Get human-readable display name for a model.
def get_model_display_name(model: BAMLModel) -> str
Usage Examples
Getting Model Configuration
from model_config import get_model_config, get_temperature
from agents.llm import BAMLModel
# Get full configuration
config = get_model_config(BAMLModel.GPT4O_MINI)
print(config) # {'temperature': 0.7}
# Get just temperature
temp = get_temperature(BAMLModel.O3_MINI)
print(temp) # 1.0 (reasoning model)
Checking Temperature Restrictions
from model_config import is_temperature_restricted
from agents.llm import BAMLModel
# Check if model has temperature restrictions
if is_temperature_restricted(BAMLModel.O3_MINI):
print("This model requires temperature 1.0")
temp = 1.0
else:
print("This model uses default temperature 0.7")
temp = 0.7
Using Display Names
from model_config import get_model_display_name
from agents.llm import BAMLModel
# Get display names for reporting
models = [
BAMLModel.GPT4O_MINI,
BAMLModel.CLAUDE_HAIKU_35,
BAMLModel.GEMINI_20_FLASH
]
for model in models:
display_name = get_model_display_name(model)
temp = get_temperature(model)
print(f"{display_name}: temperature={temp}")
# Output:
# GPT-4o Mini: temperature=0.7
# Claude Haiku 3.5: temperature=0.7
# Gemini 2.0 Flash: temperature=0.7
Benchmark Model Selection
from model_config import get_benchmark_models, get_model_display_name
# Get models for benchmarking
models = get_benchmark_models()
print("Benchmark models:")
for model in models:
print(f" - {get_model_display_name(model)}")
# Output:
# Benchmark models:
# - Devstral (OpenRouter)
# - MIMO V2 Flash (OpenRouter)
# - Nemotron Nano 12B (OpenRouter)
# ...
Custom Benchmark Configuration
from model_config import get_temperature, get_model_display_name
from agents.llm import BAMLModel
# Define custom benchmark models
custom_models = [
BAMLModel.GPT4O_MINI,
BAMLModel.GPT4O,
BAMLModel.CLAUDE_HAIKU_35,
BAMLModel.CLAUDE_SONNET_45,
BAMLModel.GEMINI_20_FLASH,
BAMLModel.DEEPSEEK_CHAT
]
# Create benchmark configuration
benchmark_config = []
for model in custom_models:
benchmark_config.append({
'model': model,
'display_name': get_model_display_name(model),
'temperature': get_temperature(model)
})
print("Benchmark configuration:")
for config in benchmark_config:
print(f" {config['display_name']}: temp={config['temperature']}")
Agent Configuration
from agents import BAMLHintGiver, BAMLGuesser
from agents.llm import BAMLModel
from model_config import get_temperature
# Create agents with appropriate temperature
model = BAMLModel.GPT4O_MINI
temp = get_temperature(model)
hint_giver = BAMLHintGiver(model)
guesser = BAMLGuesser(model)
print(f"Created agents with {get_model_display_name(model)}")
print(f"Using temperature: {temp}")
Supported Model Families
OpenAI Models
GPT-5.2 Series (December 2025)
GPT5, GPT5_MINI, GPT5_NANOGPT5_CHAT (GPT-5.2 Instant)GPT5_PRO (GPT-5.2 Pro)
GPT-4.1 Series
GPT41, GPT41_MINI, GPT41_NANO
Reasoning Models (temperature 1.0 required)
O4_MINI, O3_MINI, O3O1, O1_MINI, O1_PREVIEW
GPT-4o Series
Anthropic Models
Claude 4.5 Series (Latest)
CLAUDE_SONNET_45CLAUDE_HAIKU_45
Claude 4.x Series
CLAUDE_OPUS_41CLAUDE_SONNET_4CLAUDE_OPUS_4
Claude 3.x Series
CLAUDE_SONNET_37CLAUDE_HAIKU_35CLAUDE_HAIKU_3
Google Models
Gemini 2.5 Series
GEMINI_25_PROGEMINI_25_FLASHGEMINI_25_FLASH_LITE
Gemini 2.0 Series
GEMINI_20_FLASHGEMINI_20_FLASH_LITE
xAI Models
Grok 4 Series
GROK4GROK4_FAST_REASONINGGROK4_FAST_NON_REASONING
Grok 3 Series
GROK3, GROK3_FASTGROK3_MINI, GROK3_MINI_FAST
Other Models
DeepSeek (V3.2 - December 2025)
DEEPSEEK_CHATDEEPSEEK_REASONER
Meta Llama
OpenRouter Free Models
OPENROUTER_DEVSTRALOPENROUTER_MIMO_V2_FLASHOPENROUTER_NEMOTRON_NANOOPENROUTER_DEEPSEEK_R1T_CHIMERAOPENROUTER_DEEPSEEK_R1T2_CHIMERAOPENROUTER_GLM_45_AIROPENROUTER_LLAMA_33_70BOPENROUTER_OLMO3_32B
Temperature Guidelines
Standard Models (0.7)
Most models use temperature 0.7 for balanced creativity and consistency:
- Good for general gameplay
- Allows some randomness while maintaining quality
- Suitable for both hint giving and guessing
Reasoning Models (1.0)
OpenAI’s o-series reasoning models require temperature 1.0:
- Required by API: Cannot use other temperatures
- More deterministic despite higher value
- Optimized for reasoning tasks
Model Availability
Model availability changes frequently. The benchmark models returned by get_benchmark_models() are confirmed available as of December 2025. Check provider documentation for current availability.