# Microsoft and UC Berkeley Release TokenEconomizer

> The open-source plugin reportedly cut coding-agent token use by 18.4% to 30.9% on SWE-Bench, without changing the agent.

Oossa · 2026-10-30 · https://oossa.com/en/microsoft-and-uc-berkeley-release-tokeneconomizer

Microsoft Research and the University of California, Berkeley have released TokenEconomizer, an open-source plugin designed to lower the token use of coding agents. MarkTechPost reports that it learns from past sessions and reduced token costs by between 18.4% and 30.9% on SWE-Bench. The researchers say the plugin works without changing the agent itself.

Tokens are pieces of text an AI system processes. Using fewer can reduce the amount of model usage an agent needs, though the reported percentages are about token costs, not a stated dollar amount. The figures also do not mean every user should expect the same savings: the result is tied to SWE-Bench, and the available details do not say how the reductions varied across tasks or agent setups.

The plugin’s learning from previous sessions is the central idea described here. Rather than requiring a different coding agent, TokenEconomizer is presented as an add-on that uses session history to make token use more efficient. That could matter to teams that run coding agents repeatedly, but the description does not explain what information it stores, how it decides what to reuse, or whether users can control those choices.

## What is not yet clear

The reported range is promising as a measure of potential token savings, but the details needed to judge it are missing. MarkTechPost does not give the model or agent tested, the benchmark configuration, the number of runs, or a comparison against other token-saving methods. It also does not say whether the result was independently checked. Without those details, the percentage should be read as a reported result on one benchmark, not a guarantee for other coding work.

TokenEconomizer is described as open source, but the announcement summary does not identify a repository, license, supported agents, or setup requirements. Those details would tell developers how easily they can try it and whether it fits their tools. For now, the confirmed point is narrower: Microsoft Research and UC Berkeley have released a plugin that they say can cut coding-agent token costs on SWE-Bench without modifying the agent.

## The facts

- TokenEconomizer was released by Microsoft Research and UC Berkeley.
- The plugin is described as open source and learns from past sessions.
- MarkTechPost reports token-cost reductions of 18.4% to 30.9% on SWE-Bench.
- The plugin is described as working without changes to the coding agent.

## Why it matters

For developers who run coding agents, lower token use could reduce usage costs if similar savings carry over to their own tasks. The reported result is limited to SWE-Bench, and the available details do not establish real-world savings, dollar amounts, or which agents can use the plugin.

## Sources & references

1. [Meet TokenEconomizer: Microsoft’s Plugin That Cuts Coding Agent Costs by Up to 30.9%](https://www.marktechpost.com/2026/10/30/microsoft-tokeneconomizer-coding-agent-cost-control/) – MarkTechPost, 2026-10-30

Last updated: 2026-10-30