The authors extended the Long-Horizon-Terminal-Bench (LHTB) from 46 to 77 tasks covering 13 domains and eight task types.
Benchmark evaluates AI agents' sustained progress on long-horizon tasks with process rewards
The Long-Horizon-Terminal-Bench 2.0 extends to 77 tasks across 13 domains, uses dense progress ladders to track intermediate progress, and shows that binary outcome-only evaluation discards meaningful work.
Chinese Tech
Zongxia Li · Zhongzhi Li · Yucheng Shi · Ruhan Wang · Junyao Yang · Chengsong Huang · +11 more
University of Maryland · Tencent · University of Georgia
Research Digest··3 min read
Li et al.
Why this paper
From Tencent and 2 others
In one line
LHTB 2.0 extends agent benchmarks to 77 tasks with process rewards and three-hour evaluations.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§