microsoft/autogen · error · KeyError

Model '{model}' not found in model info

Error message

Model '{model}' not found in model info

What it means

Raised by autogen_ext.models.anthropic._model_info.get_info()/get_token_limit() as a KeyError when the model name passed to the Anthropic client is neither an exact key in the module-level _MODEL_INFO/_MODEL_TOKEN_LIMITS dicts nor a prefix match of any known base model id (matching is done on model_id.split("-2")[0]). It means AutoGen has no capability/token-limit metadata for that model. This happens when the client is constructed without an explicit model_info and the bundled registry is stale relative to the model you use.

Source

Thrown at python/packages/autogen-ext/src/autogen_ext/models/anthropic/_model_info.py:152

    "claude-3-7-sonnet-20250219": 200000,
    "claude-instant-1.2": 100000,
    "claude-2.0": 100000,
    "claude-2.1": 200000,
}


def get_info(model: str) -> ModelInfo:
    """Get the model information for a specific model."""
    # Check for exact match first
    if model in _MODEL_INFO:
        return _MODEL_INFO[model]

    # Check for partial match (for handling model variants)
    for model_id in _MODEL_INFO:
        if model.startswith(model_id.split("-2")[0]):  # Match base name
            return _MODEL_INFO[model_id]

    raise KeyError(f"Model '{model}' not found in model info")


def get_token_limit(model: str) -> int:
    """Get the token limit for a specific model."""
    # Check for exact match first
    if model in _MODEL_TOKEN_LIMITS:
        return _MODEL_TOKEN_LIMITS[model]

    # Check for partial match (for handling model variants)
    for model_id in _MODEL_TOKEN_LIMITS:
        if model.startswith(model_id.split("-2")[0]):  # Match base name
            return _MODEL_TOKEN_LIMITS[model_id]

    # Default to a reasonable limit if model not found
    return 100000

View on GitHub (pinned to 027ecf0a37)

Solutions

1. Pass explicit model_info to the client constructor (e.g. model_info={"family": ModelFamily.CLAUDE, "vision": ..., "function_calling": True, "json_output": True, "structured_output": False}) so the registry is never consulted
2. Upgrade autogen-ext to a version whose _MODEL_INFO table includes your model
3. Check the exact spelling against the keys in python/packages/autogen-ext/src/autogen_ext/models/anthropic/_model_info.py and use one of those ids (or a string starting with a registered base id before '-2')

Example fix

# before
client = AnthropicChatCompletionClient(model="claude-4-5-sonnet-2065")

# after
from autogen_core.models import ModelInfo
client = AnthropicChatCompletionClient(
    model="claude-4-5-sonnet-2065",
    model_info=ModelInfo(
        family="claude", vision=True, function_calling=True,
        json_output=True, structured_output=False,
    ),
)

from autogen_ext.models.anthropic._model_info import _MODEL_INFO

def model_supported(model: str) -> bool:
    if model in _MODEL_INFO:
        return True
    return any(model.startswith(mid.split("-2")[0]) for mid in _MODEL_INFO)

if not model_supported(MODEL):
    # supply explicit model_info instead of relying on the registry
    ...

When it happens

Trigger: Constructing AnthropicChatCompletionClient(model="claude-<new-or-typo'd-name>") without supplying model_info, or calling get_info()/get_token_limit() directly with an unlisted id. Also triggered by variant suffixes that do not share the base-id prefix before "-2" (e.g. a dated snapshot name whose prefix differs from the registry keys).

Common situations: Anthropic releases a new Claude model before autogen-ext's _MODEL_INFO table is updated; user typos the model id; user pins an old autogen-ext version but targets a newer model.

Related errors

• Required create args are missing: {required_create_args - cr
• Disallowed create args are present: {disallowed_create_args.
• model is required for AnthropicChatCompletionClient
• Failed to get login URL
• Authentication failed

AI-assisted analysis of microsoft/autogen@027ecf0a37 (2026-08-15). Data as JSON: /api/errors/5515c7b41ffbea83. Report an issue: GitHub.