Unverified50% confidenceSolutionExact time
When the primary model fails, fallback_models can shift the call to another provider, provided the fallback model supports the same tools and features (such as function calling schema, JSON mode, and context window size).
1
Sources
50%
Confidence
Long-term
Relevance
10/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn Agent scenarios, Pydantic is often used to constrain the JSON schema of model output, and a ValidationError is thrown if the model's returned content does not conform to the schema.68% similarUnverifiedMulti-model fallback in practice is not simply replacing the call address with another provider's API; different models differ in output format, degree of instruction following, tool-calling JSON structure, context window, and even token billing method, and a direct switch may cause downstream parsing failures.67% similarVerifiedUnder the tool-calling paradigm, the LLM outputs a structured function-call intent (in JSON format), which is actually executed by an external system that returns the result to the model.66% similarUnverified工具调用（Function Calling）允许模型在推理过程中调用外部函数或API，但在提示词模糊或模型版本更新后可能退回自然语言输出而非标准JSON65% similarUnverifiedAnthropic's Tool Use specification requires clients to declare available tools to the model in JSON Schema format; the model triggers calls via tool_use content blocks, and results are returned in the form of tool_result.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1113675API
curl https://kongchang.com/api/v1/knowledge/claims/1113675MCP
get_claim(id=1113675)