Tsinghua's CoACT Cuts Coding Agent Token Costs by 33 Percent

1 reports archived1 independent sourcesupdated 11/13/2026, 14:28:00

Synthesis & Latest Updates

1 Sources Cross-Validated

A team from Tsinghua University's Institute for AI proposes CoACT, an observation compression method for coding agents that optimizes for preserving the agent's next action rather than textual similarity to the original context, avoiding removal of information needed later. Across 200 SWE-bench Verified instances and three agent models, CoACT cuts total token consumption by 33.0% on average while PASS@1 moves from 69.0% to 71.0%, keeping task performance essentially intact. The work was accepted directly to ICSE 2027, a CCF-A conference.

LATEST/A team from Tsinghua University's Institute for AI proposes CoACT, an observation compression method for coding agents that optimizes for preserving the agent's next action rather than textual similarity to the original context, avoiding removal of information needed later. Across 200 SWE-bench Verified instances and three agent models, CoACT cuts total token consumption by 33.0% on average while PASS@1 moves from 69.0% to 71.0%, keeping task performance essentially intact. The work was accepted directly to ICSE 2027, a CCF-A conference.

Hourly heat curve

2 fully observed hours

Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first

机器之心 (微信公众号)T2Nov 13, 14:28·78 pts
Tsinghua's CoACT Cuts Coding Agent Token Costs by 33 Percent
- Observations account for 45.7% of total token use on SWE-bench Verified and up to 67.8% on Terminal-Bench, making them the prime compression target
- CoACT optimizes for action preservation instead of textual similarity, avoiding the risk of dropping information needed later
- It cuts token consumption by 33.0% on average with PASS@1 rising from 69.0% to 71.0%; paper and code are public
View Related Reports & Analysis Original Source