Microsoft and Berkeley open-source TokenEconomizer, cutting coding agent costs 18–31%
victor_explore · x · 2026-11-02
Microsoft Research and UC Berkeley open-sourced TokenEconomizer, a black-box plugin for coding agents like Claude Code, Codex, and GitHub Copilot CLI (4.2k stars on GitHub, Apache-2.0). The core insight: coding agents waste tokens relearning the same repo every session — which test commands work, which tools never get used.
- Micro-skills: distills what the agent would otherwise rediscover into reusable knowledge per repository
- Tool pruning: loads only the tools the repository actually uses
- Loop intervention: steps in when the agent gets stuck
Results: 18.4–30.9% lower mean cost on SWE-Bench across five agent-model combinations, cheaper than default in all 30 repository runs, with Pass@1 intact in 4 of 5 setups — and no changes to the model or the agent's code.
More from coding & agent
- As Grok bot limits tighten, open-source self-hosted agent open-duet hits 6.5k GitHub stars — matchaman11 · 2026-11-02
- Solo Dev Codes Entire Fishing Game With Opus in 3 Months, Demo Out Free — RUSuper · 2026-11-02
- Honeypot Vulnerable MCP Server Logs 3,044 Requests in 7 Days, No Real Attacks Yet — 1337Captain · 2026-11-02
- What Actually Broke When We Put Agents on Real Back-Office Work — TapNorth0888 · 2026-11-02
- Compacting earlier cuts agent costs 16-23%; GLM 5.3 costs 30x less than Grok 4.7 — yb2698 · 2026-11-02
- Codex capacity errors spike whenever OpenAI hands out banked resets — CtrlAltDwayne · 2026-11-02