OpenGPT
Search⌘K

Agent configuration · Terminal-Bench 2.1

Codex · GPT-6 Astra · medium

Historical evidence

The ranked object: model, scaffold, tools, permissions, budget, and environment together.

Compare in this benchmarkOfficial accessOpen source
Task accuracy87.0%
Cost per successful taskNot published
Published result range±1.9%

01

Configuration

Agent systemCodex · GPT-6 Astra · medium
Base modelGPT-6 Astra
Source model ID or labelGPT-6 Astra
ScaffoldCodex
Scaffold versionNot reported
ToolsTerminal shell
EnvironmentHarbor-managed terminal environments
BudgetFixed limits · at least 5 trials per task

02

Outcome

Task accuracy87.0%
Partial scoreNot published
Cost per successful taskNot published
Reported benchmark cost$619.07
Median completion timeNot published
Human interventionNot published

03

Operation

Repeat-run evidenceNot published
Published result range±1.9%
Safety noteNot published
Comparability groupterminal-bench-2-1-main
Evaluation date
Data captured

04

Trust

EvidenceListed by the benchmark publisher
SourceTerminal-Bench 2.1
Source snapshotTerminal-Bench 2.1 · publisher-verified leaderboard
Source data date
Retrieved by OpenGPT
Ranking last changed
Last checked
Tasks89 terminal tasks
Evidence levelPublisher maintained
Evidence note

The Terminal-Bench team says it ran and verified these submissions. Its 2.1 release revised tasks and the evaluation environment.

Open audit ↗

What this result does not prove

This result applies to the exact system and source conditions shown here. It does not establish a universal best agent, production reliability, data governance, or performance on a different benchmark version.

Compare like with like

Only configurations from the same benchmark version and task scope can be placed in one comparison.

Codex · GPT-6 Astra · high87.4% · GPT-6 AstraCodex · GPT-6 Astra · low86.7% · GPT-6 AstraCodex · GPT-6 Astra · max86.7% · GPT-6 AstraCodex · GPT-6 Astra · xhigh85.8% · GPT-6 Astra

Source-reported data; missing values remain unknown.

Validate before adoption

Public rankings create a shortlist. A small private trial decides whether the configuration fits your work.

  1. 1

    Choose 10–20 representative tasks and define a clear pass condition.

  2. 2

    Lock the agent, model, tools, permissions, and budget before testing.

  3. 3

    Run each important task more than once; record success, cost, time, and takeovers.

  4. 4

    Review failures and data-handling risks before a production decision.

← Agent directory and evaluations