Factory launches Router to pick cheaper AI models for coding tasks
Factory says Router is in private research preview and matched most of Opus 4.7's benchmark pass rate at 20% to 25% lower cost.
By Ryan Merket ·
Why it matters
Routing tasks to the right LLM aims to reduce coding-tool costs without giving up quality. If the preview holds up in broader use, engineering teams could keep frontier-level performance while cutting token spend.
Factory (@FactoryAI) introduced Factory Router, a model-selection system that it says automatically picks an LLM for each coding task to preserve performance while cutting token spend.
https://x.com/FactoryAI/status/2061862733126275549
Factory announced the feature in a 16-post thread on X and said Router is in private research preview in the Factory CLI and Desktop App, with more detail in a company blog post. Factory's pitch is aimed at a common AI-agent cost problem: teams often send simple fixes and documentation edits to their most expensive model because they are afraid a cheaper route will miss edge cases.
The cost claim is Factory's, not independently benchmarked here. Factory said Router reached 99% of Opus 4.7's pass rate on Terminal-Bench 2 at 20% lower cost, and 96% of Opus 4.7's pass rate on Legacy-Bench at 25% lower cost. The thread does not disclose workload mix, customer usage, or how those savings translate to a full engineering org's bill.
Factory is also selling Router as configurable, not just an automatic switchboard. It says teams can provide rules and context so model choice reflects how they operate. In replies, Factory said bring-your-own-key support is not yet available for Router, though it is working on support for popular BYOK models; it also pointed users to its model-selection docs and a recent code-review benchmark.