Atlas

Benchmarks

← All benchmarks

Dual Exam Agent Index

Indexes · 2026-10-22

Case-weighted index of the Dual Exam Agent track: 1,088 items in 15 reporting categories across physical (planning, manipulation, navigation, action sequencing), digital (web, desktop, mobile and game interaction), social (coordination and two resource-governance families) and scientific (mathematics, health, life and physical sciences) environments, mostly drawn from existing benchmarks, weighted by item count rather than by domain. Agents act under fixed environment responses; settings and harnesses differ across task families.

Top models (higher is better)

ModelScore
GPT-6 Astra31.3
GPT-5.6 Sol20.7
Gemini 3.6 Flash20.1
Qwen3.8-Max20.0
Kimi K319.8
DeepSeek V4.1 Flash18.3
GLM-5.3 Flash16.1
Doubao Seed 2.1 Pro12.7
MiniMax M311.7
MiMo-V2.511.4
Loading Atlas data…