IQ Routing
<p> Trajectory-aware LLM routing that cuts agent cost </p> <p> <a href="https://www.producthunt.com/products/iq-routing?utm_campaign=producthunt-atom-posts-feed&utm_medium=rss-feed&utm_source=producthunt-atom-posts-feed">Discussion</a> | <a href="https://www.producthunt.com/r/p/1233752?app_id=339">Link</a> </p>
Last verified:
What is IQ Routing?
IQ Routing is an API gateway that intelligently routes LLM requests to the cheapest model capable of maintaining quality across OpenAI, Anthropic, and Google endpoints. It reduces LLM spend by 40-80% through per-step model selection, caching, and session-level cost governance.
IQ Routing pricing
Pricing model: Freemium
Free tier available (card required, no charge). Specific paid tier pricing not disclosed on homepage; requires signup for details.
IQ Routing pros
- 40-80% cost reduction with no code changes to agent/RAG logic
- Works as drop-in gateway compatible with existing OpenAI/Anthropic SDKs
- Per-step routing optimizes expensive operations (planning/synthesis) vs cheap ones (verification/formatting)
- Multi-model support with vendor-lock-in safety (works inside apps requiring specific model families)
- Built-in semantic caching with ~11ms cache-hit latency, shared within teams
- Full cost transparency: per-team budgets, per-key limits, and 4-way cost breakdown (input/output/cached/thinking)
IQ Routing cons
- Additional latency introduced by routing layer (not quantified for non-cache hits)
- Quality routing classifier accuracy not independently verified
- Limited model support currently (Kimi and GLM promised but not available)
- Pricing model not detailed on marketing site (requires signup for specifics)
Frequently asked questions about IQ Routing
How much can I save with IQ Routing?
The service claims 40-80% spend reduction on measured traffic, with per-step routing selecting cheaper models where quality allows (e.g., $1.84→$0.77 for a 5-step agent loop).
How do I integrate IQ Routing?
It drops in front of existing OpenAI or Anthropic SDK/endpoints as a gateway. No changes to agent or RAG code required; can go live in 30 seconds.
Does IQ Routing work if my tool only allows one model family?
Yes. It supports model-family locking for tools like Claude Code and Cursor, routing to the right tier within that family without leaking to other vendors.
Does IQ Routing cache requests?
Yes. Semantic caching catches exact repeats and similar requests with ~11ms latency for cache hits. Cache is isolated per team.
What models does IQ Routing support?
Currently: OpenAI, Anthropic, and Google models. Kimi and GLM support is on the roadmap.