Skip to content
Hbench

Terminal-Bench 2.0

terminal-bench@2.0

89 tasks an agent carries out in a Linux terminal, each checked by its own tests.

How Terminal-Bench 2.0 measures up to the dataset checklist

Boards

Every board here runs on an open network.

The boards on Terminal-Bench 2.0
ModelHighest pass rateAttemptsHarnessesUpdated
Claude Opus 4.6
pilot82.0%74.2–88.5
5825 Oct 2026
GPT-5.3-Codex
droid77.3%69.4–84.5
5425 Oct 2026
Gemini 3.1 Pro Preview
tongagents80.2%73.5–85.8
5325 Oct 2026
GPT-5.5
Capy83.1%76.6–89.2
5225 Oct 2026
Gemini 3 Pro Preview
sageagent65.2%56.0–73.5
5225 Oct 2026
claude-opus-4-5
mux58.4%48.3–68.5
1225 Oct 2026
GLM-4.7
Claude Code33.3%25.2–41.6
5125 Oct 2026
claude-opus-4-7
vix89.9%84.7–93.9
5125 Oct 2026
gpt-5.1-codex-mini
hookele61.1%51.2–70.1
5125 Oct 2026
deepseek-v3.2
Terminus 239.6%31.5–48.1
5125 Oct 2026
Gemini 3 Flash Preview
Gemini CLI47.4%38.7–55.7
5125 Oct 2026
gpt-5-nano
spoox-o-m21.8%15.5–28.1
5125 Oct 2026
qwen3.6-35b-a3b
little-coder23.8%17.1–31.2
5125 Oct 2026
qwen3.5-9b
little-coder9.2%5.8–13.3
5125 Oct 2026
glm-5
just-another-coding-agent47.4%39.1–55.5
5125 Oct 2026
minimax-m2.7-highspeed
harness-agent42.5%33.9–50.8
5125 Oct 2026
qwen3-coder-480b-a35b-instruct-fp8
aone-agent27.2%19.8–35.1
5125 Oct 2026
claude-sonnet-4-6
simplai-agent52.1%43.6–60.7
5125 Oct 2026
gpt-5.2
mux60.7%50.6–70.8
1125 Oct 2026
Claude Sonnet 4.5
final42.7%32.6–52.8
5125 Oct 2026
MiniMax-M2.7
Indusagi Coding Agent34.4%27.4–40.9
5125 Oct 2026
GPT-5.6 Sol
Codex62.9%52.8–73.0
1124 Oct 2026
3 boards with no runs yet