IQ Routing launches trajectory-aware LLM routing

Smart model routing is crowded, but IQ Routing’s meaningful wedge is treating an agent run as a trajectory rather than a pile of independent prompts. That makes classifier accuracy, quality controls, and auditability just as important as the promised savings.

• –Per-step routing can reserve frontier models for planning and synthesis while using cheaper models for retrieval, tool calls, and verification.
• –LiteLLM already provides unified APIs, fallbacks, spend tracking, and caching; IQ Routing differentiates through session-aware routing and workload context.
• –OpenAI and Anthropic compatibility means developers can test the gateway without rewriting existing agent code.
• –Semantic caching, per-team budgets, cost ledgers, and audit logs make the product relevant to production teams, not just individual experimenters.
• –The core risk is silent quality degradation: savings only matter if the router reliably identifies which steps can tolerate a weaker model.