Skip to main content

Terminal-Bench 2.1

One harness to unlock the potential of all models

Compare Ante runs across models on the same Terminal-Bench 2.1 task set, using consistent parameters and verified benchmark results.

Best Accuracy83.9%
Leading Model
DeepSeek V4.1 Flashmax
Task Set89 tasks
Trials370 passed / 445 trials
Benchmark principlesRead more
We benchmark what we ship.Every eval uses a pinned public . No eval-only branches or benchmark-specific prompts.
The runs are auditable.Every result links its raw Harbor run, so anyone can inspect the trials behind the number.
The constraints are official.All trials follow the official Terminal-Bench parameters: 89 tasks, 5 trials per task, strict timeouts, and hardware limits.
Model org
All model orgsTB 2.1 · Same parameters: 89 tasks · 5 trials/task · Updated Oct 24, 2026
#ModelSame-modelAgentSource

Ante · TB 2.1

Ante performance

Score versus cost and duration for Ante runs. Both axes start at zero.

Ante models · cost

0%20%40%60%80%100%$0.000$0.10$0.20$0.30$0.40$0.50$0.60$0.70Cost / trial (USD)Score
The line marks the Pareto frontier among the plotted models: no run shown is both cheaper (or faster) and higher-scoring. Cost per trial is the run total divided by its trial count, so runs with different trial counts use the same unit. Trial counts and harnesses may differ; inspect each run for details. The dashed horizontal extension shows the best observed score at larger budgets, not additional measured runs.10 runs plotted, 1 hidden for missing cost data

Ante models · duration

0%20%40%60%80%100%0s10m20m30m40m50m60m70mDuration / trialScore
The line marks the Pareto frontier among the plotted models: no run shown is both cheaper (or faster) and higher-scoring. Duration is the mean per-trial duration from the raw Harbor summary, so trial counts do not change it. The dashed horizontal extension shows the best observed score at larger budgets, not additional measured runs.11 runs plotted

Terminal-Bench Reference

Verified Public Leaderboard

Same parameters: 89 tasks · 5 trials/task · Updated Sep 14, 2026 · Source: Terminal-Bench official verified rows

For how different models perform on TB 2.1, see Vals AI's Terminal-Bench 2.1 benchmark.

22 official rows
#AgentModelAccuracyRun date