Skip to content

Benchmark

TBench

Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages.

Models measured
89
Scores recorded
234
Sources
1
IQ weight
0.93

Score distribution

42.9%
Mean
42.2%
Median
22.9%
Std dev
14.2%
10th percentile
75.1%
90th percentile
0%
Min
0%91%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.93 for chat models, measured over 89 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model
101GPT 5terminal-bench-2.0, OpenHands· Terminal-Bench43.8%
102GPT-5 Codexterminal-bench-2.0, Terminus 2· Terminal-Bench43.4%
103Kimi K2.5terminal-bench-2.0, Terminus 2· Terminal-Bench43.2%
104Claude Sonnet 4.5terminal-bench-2.0, Goose· Terminal-Bench43.1%
105GPT-5.1 Codex Miniterminal-bench-2.0, Crux· Terminal-Bench43.1%
106MiniMax M2.7terminal-bench-2.0, Harness Agent· Terminal-Bench42.9%
107Claude Sonnet 4.5terminal-bench-2.0, Terminus 2· Terminal-Bench42.8%
108Claude Sonnet 4.5terminal-bench-2.0, MAYA-V2· Terminal-Bench42.7%
109MiniMax M2.5terminal-bench-2.0, cchuter· Terminal-Bench42.7%
110Claude Sonnet 4.5terminal-bench-2.0, OpenHands· Terminal-Bench42.6%
111Claude Sonnet 4.5terminal-bench-2.0, Mini-SWE-Agent· Terminal-Bench42.5%
112MiniMax M2.5terminal-bench-2.0, Terminus 2· Terminal-Bench42.2%
113GLM 5.3terminal-bench-4.0, Claude Code· Terminal-Bench41.8%
114GPT-5 Codexterminal-bench-2.0, Mini-SWE-Agent· Terminal-Bench41.3%
115GLM 5.3terminal-bench-4.1, Claude Code· Terminal-Bench41.2%
116Claude Sonnet 4.5terminal-bench-2.0, Claude Code· Terminal-Bench40.1%
117DeepSeek V4.1 Flash 1017terminal-bench-4.0, Claude Code· Terminal-Bench40%
118DeepSeek V3.2terminal-bench-2.0, Terminus 2· Terminal-Bench39.6%
119Hy 4terminal-bench-4.0, Claude Code· Terminal-Bench39.1%
120Grok 4.7terminal-bench-4.1, Grok Build· Terminal-Bench38.8%
121DeepSeek V4.1 Flash 1017terminal-bench-4.1, Claude Code· Terminal-Bench38.5%
122GPT-5.6 Solterminal-bench-4.1, Codex· Terminal-Bench38.5%
123Claude Opus 4.1terminal-bench-2.0, Terminus 2· Terminal-Bench38%
124Hy 4terminal-bench-4.1, Claude Code· Terminal-Bench37.9%
125Grok 4.7terminal-bench-4.0, Grok Build· Terminal-Bench37.6%
126GPT-5.6 Solterminal-bench-4.0, Codex· Terminal-Bench37.3%
127GLM 5.3 Flashterminal-bench-4.1, Claude Code· Terminal-Bench37%
128Claude Opus 4.1terminal-bench-2.0, OpenHands· Terminal-Bench36.9%
129GPT-5.1 Codexterminal-bench-2.0, Terminus 2· Terminal-Bench36.9%
130GLM 5.3 Flashterminal-bench-4.0, Claude Code· Terminal-Bench36.7%
131MiniMax M2.1terminal-bench-2.0, Crux· Terminal-Bench36.6%
132Kimi K2 Thinkingterminal-bench-2.0, Terminus 2· Terminal-Bench35.7%
133Claude Haiku 4.5terminal-bench-2.0, Goose· Terminal-Bench35.5%
134GPT 5terminal-bench-2.0, Terminus 2· Terminal-Bench35.2%
135Claude Opus 4.1terminal-bench-2.0, Mini-SWE-Agent· Terminal-Bench35.1%
136Claude Opus 4.1terminal-bench-2.0, Claude Code· Terminal-Bench34.8%
137GPT-5 miniterminal-bench-2.0, spoox-o-m· Terminal-Bench34.8%
138GPT-5.6 Solterminal-bench-3.0, Codex· Terminal-Bench34.6%
139GPT 5terminal-bench-2.0, Mini-SWE-Agent· Terminal-Bench33.9%
140GLM 4.7terminal-bench-2.0, Terminus 2· Terminal-Bench33.4%
141GLM 4.7terminal-bench-2.0, Crux· Terminal-Bench33.3%
142Gemini 2.5 Proterminal-bench-2.0, Terminus 2· Terminal-Bench32.6%
143GLM 5.3terminal-bench-3.0, Claude Code· Terminal-Bench32.4%
144GPT-5 miniterminal-bench-2.0, Codex CLI· Terminal-Bench31.9%
145Muse Emberterminal-bench-4.1, Muse Code· Terminal-Bench30.6%
146MiniMax M2terminal-bench-2.0, Terminus 2· Terminal-Bench30%
147Nemotron 3.5 Super VLterminal-bench-4.0, mini-swe-agent· Terminal-Bench30%
148Nemotron 3.5 Super VLterminal-bench-4.1, mini-swe-agent· Terminal-Bench30%
149Claude Haiku 4.5terminal-bench-2.0, Mini-SWE-Agent· Terminal-Bench29.8%
150GPT-5 miniterminal-bench-2.0, OpenHands· Terminal-Bench29.2%

Showing 101–150 of 234 results.

Tokens the model generated per task, as measured by the source. Comparable within this benchmark only.

Data credits: Terminal-Bench (Laude Institute), Apache-2.0 (licence)