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
1Inkling Smallterminal-bench-4.0, mini-swe-agent· Terminal-Bench91%
2Inkling Smallterminal-bench-4.1, mini-swe-agent· Terminal-Bench91%
3GPT-6 Astraterminal-bench-2.1, Codex· Terminal-Bench87.4%
4GPT 5.5terminal-bench-2.0, NexAU-AHE· Terminal-Bench84.7%
5GPT 5.5terminal-bench-2.1, Codex· Terminal-Bench83.2%
6GPT 5.5terminal-bench-2.0, Capy· Terminal-Bench83.1%
7GPT 5.5terminal-bench-2.0, Codex CLI· Terminal-Bench82.2%
8Claude Opus 4.7terminal-bench-2.0, WOZCODE· Terminal-Bench80.2%
9Gemini 3.1 Pro Previewterminal-bench-2.0, TongAgents· Terminal-Bench80.2%
10Grok 4.5terminal-bench-2.1, Cursor CLI· Terminal-Bench79.3%
11Claude Opus 4.8terminal-bench-2.1, Claude Code· Terminal-Bench78.9%
12GPT-5.6 Terraterminal-bench-2.1, Codex· Terminal-Bench78.4%
13GPT-5.3 Codexterminal-bench-2.0, SageAgent· Terminal-Bench78.4%
14GPT 5.5terminal-bench-2.1, Terminus 2· Terminal-Bench78%
15GPT-5.3 Codexterminal-bench-2.0, Droid· Terminal-Bench77.3%
16Claude Opus 4.6terminal-bench-2.0, Meta-Harness· Terminal-Bench76.4%
17Muse Spark 1.1terminal-bench-2.1, mini-SWE-agent· Terminal-Bench76.2%
18GPT-5.3 Codexterminal-bench-2.0, CodeBrain-1.5· Terminal-Bench75.8%
19GPT-5.3 Codexterminal-bench-2.0, Codelia· Terminal-Bench75.7%
20GPT-5.6 Lunaterminal-bench-2.1, Codex· Terminal-Bench75.7%
21Claude Opus 4.6terminal-bench-2.0, Capy· Terminal-Bench75.3%
22GPT-5.3 Codexterminal-bench-2.0, Simple Codex· Terminal-Bench75.1%
23Gemini 3.1 Pro Previewterminal-bench-2.0, Terminus-KIRA· Terminal-Bench74.8%
24Claude Opus 4.6terminal-bench-2.0, Terminus-KIRA· Terminal-Bench74.7%
25GPT-5.3 Codexterminal-bench-2.0, Mux· Terminal-Bench74.6%
26Gemini 3 Pro Previewterminal-bench-2.1, Terminus 2· Terminal-Bench73.9%
27Claude Opus 4.6terminal-bench-2.0, MAYA-V2· Terminal-Bench72.1%
28Claude Opus 4.6terminal-bench-2.0, TongAgents· Terminal-Bench71.9%
29GPT-5.3 Codexterminal-bench-2.0, spoox-o-m· Terminal-Bench71.5%
30Claude Opus 4.6terminal-bench-2.0, Droid· Terminal-Bench69.9%
31Gemini 3 Pro Previewterminal-bench-2.0, Ante· Terminal-Bench69.4%
32GPT-5.3 Codexterminal-bench-2.0, IndusAGI Coding Agent· Terminal-Bench69.1%
33Claude Opus 4.7terminal-bench-2.1, Claude Code· Terminal-Bench68.9%
34Claude Opus 4.6terminal-bench-2.0, Crux· Terminal-Bench66.9%
35Claude Opus 4.6terminal-bench-2.0, Mux· Terminal-Bench66.5%
36GPT-5.2 Codexterminal-bench-2.0, Deep Agents· Terminal-Bench66.5%
37Claude Opus 4.7terminal-bench-2.1, Terminus 2· Terminal-Bench66.1%
38GPT 5.5terminal-bench-2.0, clnkr· Terminal-Bench66.1%
39Gemini 3 Pro Previewterminal-bench-2.1, Gemini CLI· Terminal-Bench65.8%
40Gemini 3.1 Pro Previewterminal-bench-2.1, Gemini CLI· Terminal-Bench65.8%
41Claude Opus 5.5terminal-bench-4.1, Claude Code· Terminal-Bench65.8%
42Gemini 3.1 Pro Previewterminal-bench-2.1, Terminus 2· Terminal-Bench65.6%
43Gemini 3 Pro Previewterminal-bench-2.0, SageAgent· Terminal-Bench65.2%
44GPT 5.2terminal-bench-2.0, Droid· Terminal-Bench64.9%
45Claude Opus 5.5terminal-bench-4.0, Claude Code· Terminal-Bench64.9%
46GPT-5.3 Codexterminal-bench-2.0, Terminus 2· Terminal-Bench64.7%
47Gemini 4 Carbonterminal-bench-4.1, Antigravity CLI· Terminal-Bench64.6%
48Gemini 3 Flashterminal-bench-2.0, Junie CLI· Terminal-Bench64.3%
49Claude Opus 4.5terminal-bench-2.0, Droid· Terminal-Bench63.1%
50Claude Opus 4.6terminal-bench-2.0, Terminus 2· Terminal-Bench62.9%

Showing 1–50 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)