Back

Fig.inc

Ghost
fig.inc· irregular· Published 10 days ago

Fig.inc publishes research-blog-style posts from its Manifold Research collaboration on evaluating frontier AI agents, covering maze-navigation failures, open-ended data for computer-use agents, and jagged task performance across web and physical domains. Readers get concrete benchmark results, released datasets, and analysis of how and why models fail.

ai agentsfrontier modelsbenchmarksosworldmaze evaluationmultinetembodied aicomputer useagentic tasksmodel evaluationtraining dataautonomous agents
Platform
Ghost
Frequency
Irregular
Latest issue
Oct 13, 2026
Language
English
Written for
AI researchers, engineers and practitioners evaluating agent systems
Tone
analytical
Tracked since
Oct 22, 2026

Website: https://www.fig.inc/