- 官方分层与路由主轴：OpenAI
A practical guide to building agents、OpenAIGPT-4.1、AnthropicBuilding effective agents、GoogleGenerationConfig/Google models、MicrosoftModel router、AWSIntelligent Prompt Routing - 路由与级联实证主轴：
FrugalGPT、Hybrid LLM、RouteLLM、RouterBench、A Unified Approach to Routing and Cascading for LLMs、Adaptive LLM Routing under Budget Constraints、Firewall Routing、RouterEval - 小模型与蒸馏主轴：
DistilBERT、TinyBERT、Phi-3 Technical Report、OpenAIModel Distillation in the API、OpenAIStructured Outputs - 服务经济学主轴：
PagedAttention、Orca、DistServe、Sarathi-Serve、NVIDIAInference Optimization、OpenAIPrompt Caching、AnthropicPrompt caching - 升级条件与评测主轴：
Selective Classification、Language Model Cascades: Token-Level Uncertainty and Beyond、Optimising Calls to Large Language Models with Uncertainty-Based Two-Tier Selection、AWSprompt routing and prompt caching、AnthropicDemystifying evals for AI agents - 企业实践主轴：LinkedIn
Musings on Building a Generative AI Product、SpotifyOur Multi-Agent Architecture for Smarter Advertising、GoogleContext caching overview、Anthropic / AppFolio、Anthropic / Grafana
附录 G：分章资料清单