BerriAI/litellm · error · ValueError

Model {model_info['key']} does not have 'input_cost_per_char

Error message

Model {model_info['key']} does not have 'input_cost_per_character' or 'input_cost_per_token'

What it means

select_cost_metric_for_model decides between character-based and token-based cost math by inspecting model_info; it requires at least one of 'input_cost_per_character' or 'input_cost_per_token'. When a model entry defines only other cost fields (e.g. only output costs, or only cost_per_second), the helper raises this ValueError because it cannot choose a metric for the input side.

Source

Thrown at litellm/litellm_core_utils/llm_cost_calc/utils.py:114

    Subtracts cached tokens from prompt tokens if applicable.
    """
    details: Final = _parse_prompt_tokens_details(usage)
    return usage.prompt_tokens - details["cache_hit_tokens"]


def select_cost_metric_for_model(
    model_info: ModelInfo,
) -> Literal["cost_per_character", "cost_per_token"]:
    """
    Select 'cost_per_character' if model_info has 'input_cost_per_character'
    Select 'cost_per_token' if model_info has 'input_cost_per_token'
    """
    if model_info.get("input_cost_per_character"):
        return "cost_per_character"
    elif model_info.get("input_cost_per_token"):
        return "cost_per_token"
    else:
        raise ValueError(
            f"Model {model_info['key']} does not have 'input_cost_per_character' or 'input_cost_per_token'"
        )


def _generic_cost_per_character(
    model: str,
    custom_llm_provider: str,
    prompt_characters: float,
    completion_characters: float,
    custom_prompt_cost: float | None,
    custom_completion_cost: float | None,
) -> tuple[float | None, float | None]:
    """
    Calculates cost per character for aspeech/speech calls.

    Calculates the cost per character for a given model, input messages, and response object.

    Input:

View on GitHub (pinned to 6c2dcb801b)

Solutions

1. Add input_cost_per_token (or input_cost_per_character for character-billed models) to the model's model_info in your config.yaml
2. If using a custom model_prices JSON, add the input cost fields (0.0 is valid for free models) and reload
3. Upgrade LiteLLM so the bundled pricing DB covers the model if it is a known public model
4. Verify with litellm.get_model_info(model) that the resolved entry actually has the fields you set (wildcard/deployment overrides can shadow it)

Example fix

# before (config.yaml)
model_info:
  output_cost_per_token: 0.00001   # missing input cost -> ValueError

# after
model_info:
  input_cost_per_token: 0.0000025
  output_cost_per_token: 0.00001

from litellm.litellm_core_utils.llm_cost_calc.utils import select_cost_metric_for_model

def pricing_complete(model_info: dict) -> bool:
    return bool(model_info.get('input_cost_per_character') or model_info.get('input_cost_per_token'))

info = litellm.get_model_info(model)
assert pricing_complete(info), f"model '{model}' lacks input_cost_per_token; cost calc will raise"

def has_input_cost(model_info: dict) -> bool:
    return bool(model_info.get('input_cost_per_character') or model_info.get('input_cost_per_token'))

When it happens

Trigger: A cost calculation path that uses this selector (character/token cost utils) running against a model whose model_info — from model_prices_and_context_window.json, a custom pricing file, or config.yaml model_info — lacks both input_cost_per_character and input_cost_per_token (e.g. only output_cost_per_token set, or a free model declared with {} pricing).

Common situations: Hand-written custom pricing entries that forget input costs; free/community endpoints declared with no input cost; copy-paste of an output-only pricing block; older pricing entries missing the character fields used by character-billed models.

Related errors

• [Non-Blocking] LiteLLM.Success_Call Error: {e}
• Error: {response.status_code} - {response.text}
• Missing Authorization header
• Invalid bearer token
• Invalid API key

AI-assisted analysis of BerriAI/litellm@6c2dcb801b (2026-08-15). Data as JSON: /api/errors/d7b893a166cd6888. Report an issue: GitHub.