Terminal-Bench 2.1
NAESTRO / benchmark archive
Historical local snapshot over the full 89-task TB2.1 set — Harbor adapter, xhigh effort.
Evidence scopeHistorical local snapshot. Self-reported retained results with task-level rows and filters preserved for inspection. This page does not independently verify the run or claim current leaderboard standing.
Official TB2.1 leaderboard ↗ · checked September 15, 2026.
GPT-6 Astra release: September 3, 2026 · 0.4B tokens · $0.8k. Error range: 95% confidence interval.
Different runs and evaluation conditions; this is context, not a controlled head-to-head comparison.