Repository navigation
Releases: Mtrya/AstroAgentBench
Releases · Mtrya/AstroAgentBench
Release list
AstroAgentBench v2.0.0
AstroAgentBench v2.0.0 publishes the benchmark in Harbor-native formats for reproducible agent evaluation.
- RL Environment taskset: 131 tasks across seven benchmark configurations, with native Harbor loading and dataset viewing.
- Agent traces: 115 Claude Code and Codex sessions rendered by Hugging Face's native trace viewer. The source archive is preserved on Zenodo.
- Runtime image:
ghcr.io/mtrya/astroagentbench-python@sha256:fe9733cad55519aa08dd0db9abb51444051ef5f53d6ac1ef0b4b6592491437d1.
All seven oracle smoke checks passed against the promoted Hugging Face taskset. Publication evidence: taskset workflow and traces workflow.