Act on this, mostly signal

Token Beggar routes coding subtasks to cheaper models

What to do

Install the Token Beggar Claude Code plugin if large subtasks drive your agent bill.

Why

The author reports 1.42 times lower cost on 18 bug fixes with a 69 percent pass rate versus 70 percent baseline. All figures are self reported.

Read the source at github.com

What the research found

Token Beggar v0.3.0 is a Claude Code plugin that routes large subtasks to a cheaper worker model; the author reports 1.42x lower cost on 18 real bug fixes with hidden-test pass rate 69% versus 70% baseline, all self-reported.

Claims and how well the evidence supports them

Token Beggar is a Claude Code plugin that routes large subtasks to a cheaper model 96%
On 18 real bug fixes it cost 1.42x less than working alone 98%
Pass rate on hidden tests was 69% with Token Beggar versus 70% without 98%
It needs Claude Code and Python 3.10 or newer and uses only the standard library 99%

Evidence

github.com https://github.com/Arthur031221/Token-Beggar

Actionability rated medium.