import re
from pydantic import BaseModel
class ChatModelCostPer1MToken(BaseModel):
input: float
cached: float | None = None
output: float
class ChatModelCostTable(BaseModel):
costs: dict[str, ChatModelCostPer1MToken] = {
"gpt-4o-mini": ChatModelCostPer1MToken(input=0.15, cached=0.075, output=0.6),
"gpt-4o": ChatModelCostPer1MToken(input=2.5, cached=1.25, output=10.0),
"gpt-4.5-preview": ChatModelCostPer1MToken(input=75.0, cached=37.5, output=150.0),
"gpt-4.1-nano": ChatModelCostPer1MToken(input=0.1, cached=0.025, output=0.4),
"gpt-4.1-mini": ChatModelCostPer1MToken(input=0.4, cached=0.1, output=1.6),
"gpt-4.1": ChatModelCostPer1MToken(input=2.0, cached=0.5, output=8.0),
"o1-mini": ChatModelCostPer1MToken(input=3.0, cached=0.55, output=12.0),
"o1-preview": ChatModelCostPer1MToken(input=15.0, output=60.0),
"o1": ChatModelCostPer1MToken(input=15.0, cached=7.5, output=60.0),
"o1-pro": ChatModelCostPer1MToken(input=150.0, output=600.0),
"o3-mini": ChatModelCostPer1MToken(input=1.1, cached=0.55, output=4.4),
"o3": ChatModelCostPer1MToken(input=10.0, cached=2.5, output=40.0),
"o4-mini": ChatModelCostPer1MToken(input=1.1, cached=0.275, output=4.4),
}
def get_cost(self, model_name: str) -> ChatModelCostPer1MToken | None:
return self.costs.get(model_name, None)
def list_models(self) -> list[str]:
return list(self.costs.keys())
def get_openai_model_name_from_version(model_version: str) -> str:
return re.sub(r"-\d{4}-\d{2}-\d{2}$", "", model_version)
def calculate_cost(input_tokens: int, cached_tokens: int, output_tokens: int, model_version: str) -> float:
model_name = get_openai_model_name_from_version(model_version)
cost_per_1m_tokens = ChatModelCostTable().get_cost(model_name)
if cost_per_1m_tokens is None:
print(f'Model name "{model_name}" is not found in the cost table.')
return 0.0
input_cost_per_1m_tokens = cost_per_1m_tokens.input / 10**6
cached_cost_per_1m_tokens = cost_per_1m_tokens.cached / 10**6 if cost_per_1m_tokens.cached is not None else 0.0
output_cost_per_1m_tokens = cost_per_1m_tokens.output / 10**6
non_cached_input_tokens = input_tokens - cached_tokens
cost = (
non_cached_input_tokens * input_cost_per_1m_tokens
+ cached_tokens * cached_cost_per_1m_tokens
+ output_tokens * output_cost_per_1m_tokens
)
return cost
