Smart Model Routing

The SmartModelRouter automatically selects the optimal LLM model based on task complexity, saving costs by routing simple tasks to cheaper models.

How It Works

1. Analyze the prompt to classify complexity (trivial/simple/moderate/complex/expert)
2. Route to the appropriate model tier
3. Fall back to a higher tier if the selected model fails

"fix typo" ──► trivial ──► gpt-5.4-mini ($0.15/M)
"add button" ──► simple ──► gpt-5.4-mini ($0.15/M)
"refactor auth" ──► complex ──► claude-sonnet-4-6 ($3/M)
"design system" ──► expert ──► claude-opus-4-6 ($15/M)

Usage

import { SmartModelRouter } from '@opensin/sdk'

const router = new SmartModelRouter({
  models: {
    trivial: 'gpt-5.4-mini',
    simple: 'gpt-5.4-mini',
    moderate: 'gpt-5.4',
    complex: 'claude-sonnet-4-6',
    expert: 'claude-opus-4-6',
  },
  fallbackOnError: true,
})

// Automatic selection
const model = router.selectModel('Fix the typo in README.md')
// Returns: 'gpt-5.4-mini'

const model2 = router.selectModel('Refactor the entire authentication system to use JWT')
// Returns: 'claude-sonnet-4-6'

Complexity Classification

The router uses keyword analysis and prompt structure to classify tasks:

Level	Indicators	Examples
Trivial	typo, rename, comment	"Fix typo in line 5"
Simple	add, update, format	"Add a loading spinner"
Moderate	implement, create, build	"Implement pagination"
Complex	refactor, redesign, migrate	"Refactor auth to OAuth2"
Expert	architect, design system, security audit	"Design the plugin system"

Fallback Chain

When a model fails (rate limit, error, insufficient output), the router automatically escalates:

gpt-5.4-mini → gpt-5.4 → claude-sonnet-4-6 → claude-opus-4-6

const router = new SmartModelRouter({
  models: { ... },
  fallbackOnError: true,
  maxRetries: 2,
  fallbackChain: ['gpt-5.4-mini', 'gpt-5.4', 'claude-sonnet-4-6', 'claude-opus-4-6'],
})

Cost Savings

Typical cost reduction with smart routing vs. always using the best model:

Workload	Without Routing	With Routing	Savings
100 tasks (mixed)	$4.20	$0.95	77%
Code review (simple)	$1.50	$0.15	90%
Architecture (expert)	$7.50	$7.50	0%

Smart routing saves the most when your workload has a mix of simple and complex tasks, which is the typical case for coding agents.

Integration with Agent Loop

const agent = new AgentLoop({
  model: router,  // pass router instead of model string
  tools: toolRegistry,
})

The agent loop calls router.selectModel(prompt) on each turn, so the model can change mid-conversation as task complexity evolves.

---

Relevante Mandate

Mandat	Priority	Regel
Bun-Only	-1.5	bun install / bun run statt npm
Annahmen-Verbot	-5.0	KEINE Diagnose ohne Beweis
Test-Beweis-Pflicht	0.0	KEIN "Done" ohne echten Test-Lauf

→ Alle Mandate