METR (reported by MIT Technology Review) · Time horizon benchmark
METR time horizon plot update
METR has updated its time horizon benchmark, which measures how long a task a model can complete relative to human completion time. The new suite adds longer tasks and more open-ended ones, such as designing AI evaluations and scoring philosophical arguments. Models fall short of human performance on these judgment-heavy tasks, though newer models such as Opus 5.5 score "borderline above the bar" on evaluation design, up from "clearly below the bar" a year earlier. Many completion times are estimates rather than measured human results.
Open the source ↗Similar Research & Safety releases
- MedChemBench
OpenAI · Medicinal chemistry benchmark · Week 48, 2026
- StrideWorld
Chuang Wang, Rui Nie, Jie Song et al. (Shanghai Jiao Tong University) · World Model Benchmark · Week 48, 2026
- Self-Improving Memory
Meta FAIR · Self-Improving Memory · Week 48, 2026
- TuringSpace: Grounding Spatial Intelligence in Action with Interactive CAPTCHAs
Meta (Xuwei Ding et al.) · Spatial Action Benchmark · Week 48, 2026
- Game2Player
Xirui Jiang, Yuxi Chen, Rui Yang, Feng Luo, Pengcheng Wang, Bingxuan Li, Zihao Zhao, Yushan Li (University of Illinois at Urbana-Champaign) · Gameplay Experience Scaling · Week 48, 2026
- AscendGYM
SLAI-AITP (AI Training Platform Team) · Kernel Optimisation Benchmark · Week 48, 2026
Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.