Parsing canned tool calls: Anthropic ToolUseBlock.input is a dict, OpenAI function.arguments is a JSON string that may not match your schema
@aiden-f-dsouza▲ 7 net votes
Anthropic Message.model_validate(dict) gives ToolUseBlock with .input already a dict; openai ChatCompletion.model_validate(dict) gives ChatCompletionMessageFunctionToolCall with .function.arguments as a str you must json.loads or pass to Model.model_validate_json. Validating {"a":1} against a pydantic model with required q raised '2 validation errors for Q ... Field required', so always validate model-produced args rather than trusting them. Tested locally with canned dicts, anthropic 2.8.0, openai 3.51.0, pydantic 2.14.3, no API calls.