SCOI Atlas · L2 Models
L2D Model Routing & Composition
Cost-quality routers that pick the right model per request; mixture-of-experts style composition.
TL;DR · Direct answer
L2D Model Routing & Composition is a sublayer of L2 Models in the Supply Chain of Intelligence™ (SCOI) by Anand Arivukkarasu. It is a required capability but not on its own a durable moat. Routing is a margin lever, not a moat. But without it, your gross margin will trail competitors by 10–20 points.
What actually matters at L2D
- Routing can cut inference cost 40–70% at fixed quality — but requires an eval regime.
- Naive routers regress on tail queries; measure P95, not average.
- Composition is where speculative decoding, small-first, cascade-then-large patterns live.
The startup lens
Routing is a margin lever, not a moat. But without it, your gross margin will trail competitors by 10–20 points.
Vertical lens — how this plays across categories
AI-native SaaS
Router with per-tenant cost/quality thresholds.
Consumer AI
Small-model-first cascade; only escalate on confidence miss.
Agent platforms
Route per-tool, per-task, not per-conversation.
How to defend L2D
- Eval-driven routing decisions.
- Router that improves as L1D outcome data grows.
Other sublayers in L2 Models
L2A
Foundation & Multimodal Models
Frontier models from OpenAI, Anthropic, Google, Meta, xAI, Mistral, DeepSeek and their open-weight peers.
L2B
Specialized & Fine-Tuned Models
Domain-tuned models — code, biology, legal, finance, radiology, robotics.
L2C
Embedding & Retrieval
Text, image and code embeddings; rerankers; hybrid search.
L2E
Reasoning & World Models
Chain-of-thought, tree search, o-series style reasoning, and true world models for physical AI.