Microsoft and UC Berkeley release TokenEconomizer for coding agents
Original title: Meet TokenEconomizer: Microsoft’s Plugin That Cuts Coding Agent Costs by Up to 30.9%
On 277 SWE-Bench tasks it cut mean cost 18.4% to 30.9% across five agent-model settings without changing weights.
Researchers from Microsoft Research and UC Berkeley released TokenEconomizer, an open-source plugin that learns from past repository sessions to reduce later token spend. It works with Claude Code, Codex, GitHub Copilot CLI and Pi, and does not modify model weights or agent code. Across five agent-model settings on 277 SWE-Bench tasks, mean cost fell 18.4% to 30.9%, and Pass@1 rose in four of five settings. Savings were smaller on ReposVul audits, at 6.4% to 9.3%, where Pass@1 dropped 1.0 point on Opus 4.8.
- Best reported result was 30.9% lower cost on Claude Code with Opus 4.8, or 33.2% combined with RTK.
- It needs Python 3.10+ and no GPU or API key of its own.
- Three mechanisms are micro-skills, tool pruning and failure-loop intervention.
- Costs also fell 16.0% to 21.2% on Web-Bench and 16.5% to 19.6% on CredData.
- Vendors
- Microsoft
- Organizations
- Microsoft ResearchUC Berkeley
From the original article
8 min read at marktechpost.comMicrosoft Research and UC Berkeley released TokenEconomizer, an open-source plugin that learns from past sessions to cut coding agent token costs by 18. 4% to 30. 9% on SWE-Bench, with no change to the agent.
By Asif Razzaq · marktechpost.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.