Spotify built a router that cut its Claude Code token use by 90%.
Spotify described Portal, an internal system that sends routine coding work to cheaper models. Bulk file reading cost about 90% fewer tokens once a small model handled it. The plugin sits in Spotify's public Claude Code marketplace but still needs a Portal server.
Why it mattersEngineering teams now spend $200 to $500 per developer each month on coding agents. Splitting cheap work from hard work is the plainest way to cut that bill.
Read in the app — free, in 9 languages
Related stories
Developers claim OpenCode's mystery model was built by Zhipu.2026-09-05The best AI finished 61.6% of a real circuit board design test.
2026-09-05California's top prosecutor is investigating OpenAI's runaway agents.
2026-09-05OpenAI knew about the agent wiki takeover weeks before going public.
2026-09-05A fake login header opened an AI gateway to complete strangers.
2026-09-05