Supported Model Catalog

CTS Model Router provides access to both frontier cloud models and high-efficiency open-weight models via unified aliases.

---

1. Universal Model Aliases

You can target specific models or use dynamic routing aliases that automatically select the best model based on latency and cost:

Alias / Model ID	Provider Family	Max Context	Typical TTFT	Primary Use Case
auto-fast	Dynamic (Gemini Flash / DeepSeek)	128k	~180ms	High-volume APIs, summarization, extraction
auto-reasoning	Dynamic (DeepSeek-R1 / Claude 3.7)	128k	~350ms	Code analysis, mathematical proofs, logic
deepseek-r1-distill	DeepSeek / Hosted vLLM	64k	~210ms	Competitive reasoning at commodity pricing
claude-3-7-sonnet	Anthropic Claude	200k	~400ms	SOTA coding, nuanced reasoning, vision
gpt-4o	OpenAI	128k	~380ms	General multimodal intelligence
gemini-2-5-flash	Google Cloud	1,000k	~160ms	Massive context retrieval, video analysis

---

2. Dynamic Fallback Rules

Configure failover policies directly in your request payload using custom headers or parameters:

{
  "model": "claude-3-7-sonnet",
  "fallback_models": ["gpt-4o", "deepseek-r1-distill"],
  "timeout_ms": 2000
}

If the primary model exceeds 2,000ms or returns an upstream error, the router transparently falls back to the specified alternatives.