TokenEconomizer Reports Lower Coding-Agent Costs Across Benchmarks
Technology · First reported 29 Oct, 04:53 · Updated 05:45
Microsoft Research and UC Berkeley researchers introduced TokenEconomizer, a plugin that learns from prior repository sessions to reduce coding-agent costs. In reported tests, it lowered mean SWE-Bench costs by 18.4% to 30.9% across five agent-model settings.
- TokenEconomizer learns from completed sessions in a repository and adjusts prompts, available tools, and responses to repeated failures; it does not modify model weights or agent code.
- Across five SWE-Bench agent-model settings, the article reports mean cost reductions of 18.4% to 30.9%; costs were lower in all 30 repository-setting cells.
- Pass@1 improved in four of five settings and fell by one point for Claude Code with Opus 4.8; ReposVul savings were smaller, at 6.4% to 9.3%.
Summary written by Lux AI from 2 articles
Key facts
- Researchers
- Microsoft Research and UC Berkeley
- Reported SWE-Bench cost reduction
- 18.4% to 30.9% across five agent-model settings
- Repository-setting cells with lower cost
- 30 of 30
- Pass@1 result
- Improved in four of five settings; declined by one point on Claude Code with Opus 4.8
- Supported agents
- Claude Code, Codex, GitHub Copilot CLI, and Pi
Source spectrum
Far left
Left
Lean left
Centre
Lean right
Right
Far right
Unrated
Left 0 · Centre 1 · Right 0 of 1 rated sources · +1 unrated
All coverage
Coverage over time
LeftCentreRight
Articles per hour since first reported, by side.
Ownership
- Corporation 50% (1)
- Unknown 50% (1)
Share of the 2 sources by owner type.