​
Overview

Adaptive’s intelligent routing system analyzes every request and automatically selects the optimal AI model and provider based on content, complexity, cost, and performance requirements. This happens transparently - you send requests and get back the best possible responses at the lowest cost.

​
How It Works

1

Classification

Neural classifiers analyze your prompt for task type (11 types), domain (25 domains), complexity, and reasoning requirements
2

Route Selection

Routes to premium models for complex tasks or specialist models for simple domain-specific tasks
3

Model Selection

Selects optimal model using task rankings for premium models or domain-task matrix for specialist models

​
Active Providers

ProviderModelsStrengthsBest For
OpenAIgpt-4o, gpt-4o-mini, o1, o3General purpose, function callingBalanced tasks, production workloads
DeepSeekdeepseek-chat, reasonerCode generation, math reasoningProgramming, technical analysis
Grokgrok-3, grok-3-miniUltra-fast inference, creativityReal-time apps, creative writing
GroqFast inference modelsExtremely low latencyHigh-throughput applications

​
Routing Examples

​
Simple Q&A → gpt-4o-mini

“What’s the capital of France?” → 92% cost savings

​
Code Generation → deepseek-chat

“Write a Python binary search function” → 78% cost savings

​
Creative Writing → grok-3

“Write a marketing story about sustainable tech” → 65% cost savings

​
Controlling Routing

​
Cost Bias Parameter

Control the cost vs performance trade-off: 
const completion = await openai.chat.completions.create({
  model: "",
  messages: [{ role: "user", content: "Explain machine learning" }],
  cost_bias: 0.2, // 0 = cheapest, 1 = best performance
});
  • 0.0-0.3: Maximum savings (cheapest models)
  • 0.4-0.6: Balanced (default)
  • 0.7-1.0: Maximum quality (best models)

​
Provider Constraints

Limit which providers can be selected: 
const completion = await openai.chat.completions.create({
  model: "",
  messages: [{ role: "user", content: "Write code to sort an array" }],
  provider_constraints: ["deepseek", "anthropic"],
});

​
Performance

  • 94% accuracy in optimal model selection
  • 97% user satisfaction with routing decisions
  • 2ms latency for routing decisions
  • 99.9% uptime across all providers

​
Transparency

Every response includes the selected provider: 
{
  "id": "chatcmpl-123",
  "model": "gemini-2.0-flash",
  "provider": "gemini",
  "choices": [...],
  "usage": {...}
}
Monitor routing decisions in your dashboard with provider distribution, cost savings, and performance analytics.

​
FAQ