Skip to content
IQ Routing
IQ Code documentation

IQ Code

Your cost, cache clock and plan pace in Claude Code. Optional routing when you want it.

Public release 1.0.0 · For the terminal · Tested on macOS and Linux; Windows needs Git for Windows · The desktop app's Code tab is not verified yet

Start with the local tools

Free includes the API-equivalent cost meter, prompt cache countdown, 5-hour and weekly plan pace, budgets, model and thinking labels and the IQ look. Open /iq for Home, Routing, Cost, Plan and Settings.

Routing starts off in every session. IQ Code routing needs the Personal plan ($6/month, 2,000 routing decisions a month), which also adds rough session cost estimates. Your Claude usage stays on your Anthropic plan.

A routing decision is one request to IQ, made for each new prompt you send while routing is on and for each subagent task Claude starts in that session (not forks, workflow agents or teammates). Subagent-heavy sessions use decisions faster.

See IQ Code · Compare plans

Source and license · Report an issue. For account and billing questions, email support@iq-routing.com.

Session cost estimates

Personal shows a rough estimate of the whole session's total cost in Opus- and Sonnet-led sessions without subagents or compaction, after the first call and up to 64 calls. A typical estimate lands within about 2.4x of the final session cost. Measured cost is always shown. Paying Team and Enterprise plans also include these estimates.

Estimates are requested only while routing is on. With a usable IQ key, IQ Code reads your dashboard's session defaults from IQ at startup and at most once every ten minutes, even with routing off. That request is a GET with no body, and it sends no prompt content. Without a usable IQ key, it sends nothing to IQ. Cost estimate requests send usage numbers (conversation and cache-read tokens), model and thinking-level labels, call, helper and compaction counts, spend, active time, lifecycle and usage-status labels, a random session number and a request sequence number to IQ. They contain no prompt, answer, file or tool text.

Session estimates use a model fit in part on TraceLab v0.0.2 (SyFI Lab, University of Washington), CC BY 4.0. Dataset source · Dataset attribution. Changes: retained only numeric and structural metadata; converted records to job/call views; repriced recorded token buckets using the 2026-10-06 rate card; fitted population and user-shrinkage cost priors. Session boundaries substitute for verified task boundaries. No endorsement is implied.