Terminal-Bench 2.0 scores agents on completing real tasks in a command-line environment; it is hosted by Stanford, Harbor and the Laude Institute (leaderboard). It enters the corpus as the benchmark GPT-5.2-Codex topped on release, then becomes the clearest scoreboard for lab-versus-lab velocity when Claude Opus 4.6’s record fell to GPT-5.3-Codex in under thirty minutes. In June 2026 it is the yardstick on which an agent that rewrote its own scaffolding gained double digits with no human engineers involved.
Benchmark
Terminal Bench 2.0
Agentic terminal-task benchmark whose record changed hands between labs within half an hour and which a self-rewriting harness later lifted by double digits.
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to December 19, 2025 (theinnermostloop.substack.com)
Development timeline
Newest reporting dates first. Reporting dates may differ from event dates.
Knowledge graphopen in browser →
Published by
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.