- 官方选型框架主轴：OpenAI
Models/All models、AnthropicChoosing the right model、GoogleVertex AI Model Optimizer/Google models、Microsoftmodel benchmarks/deployment types、AWSmodel catalog - 多维能力向量主轴：
HELM、BIG-bench、BBH、LongBench、T-Eval、BIRD、HarmBench - 路由、级联与多模型编排主轴：
FrugalGPT、LLM Cascades with Mixture of Thought Representations、RouteLLM、AWSIntelligent Prompt Routing、MicrosoftModel router - 成本、延迟与系统约束主轴：OpenAI
Prompt Caching、OpenAILatency optimization、Orca、PagedAttention、DistServe - 风险治理与高风险选型主轴：
EU AI Act、NIST AI 600-1、GPT-4o System Card、Claude 4 System Card、Larger and more instructable language models become less reliable、Must Be Taught to Know What They Don’t Know、Learn to Refuse、Robots That Ask For Help - 系统补强与业务适配主轴：Anthropic
Building effective agents、OpenAIA practical guide to building agents、OpenAIStructured Outputs、Googleadvanced governance capabilities for Vertex AI Agent Builder、OpenAIInside OpenAI’s in-house data agent、RAG、Toolformer、ReAct
附录 G：分章资料清单