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
151MiniMax M2.1terminal-bench-2.0, Terminus 2· Terminal-Bench29.2%
152Claude Haiku 4.5terminal-bench-2.0, Terminus 2· Terminal-Bench28.3%
153Mistral Large 4terminal-bench-4.0, Claude Code· Terminal-Bench27.9%
154Kimi K2 Instructterminal-bench-2.0, Terminus 2· Terminal-Bench27.8%
155Mistral Large 4terminal-bench-4.1, Claude Code· Terminal-Bench27.6%
156Claude Haiku 4.5terminal-bench-2.0, Claude Code· Terminal-Bench27.5%
157Grok 4terminal-bench-2.0, OpenHands· Terminal-Bench27.2%
158Qwen3 Coderterminal-bench-2.0, Dakou Agent· Terminal-Bench27.2%
159MiMo V2.6 Proterminal-bench-4.0, mini-swe-agent· Terminal-Bench27%
160MiMo V2.6 Proterminal-bench-4.1, mini-swe-agent· Terminal-Bench27%
161Qwen3.8 Maxterminal-bench-4.0, Claude Code· Terminal-Bench27%
162Qwen3.8 Maxterminal-bench-4.1, Claude Code· Terminal-Bench27%
163Kimi K2 Instructterminal-bench-2.0, OpenHands· Terminal-Bench26.7%
164Grok 4.6terminal-bench-3.0, Grok Build· Terminal-Bench26.5%
165Gemini 2.5 Proterminal-bench-2.0, Mini-SWE-Agent· Terminal-Bench26.1%
166Grok Code Fast 1terminal-bench-2.0, Mini-SWE-Agent· Terminal-Bench25.8%
167Grok 4terminal-bench-2.0, Mini-SWE-Agent· Terminal-Bench25.4%
168Qwen3 Coderterminal-bench-2.0, OpenHands· Terminal-Bench25.4%
169Qwen3.6 35B A3Bterminal-bench-2.0, little-coder· Terminal-Bench24.6%
170GLM 4.6terminal-bench-2.0, Terminus 2· Terminal-Bench24.5%
171GPT-5 miniterminal-bench-2.0, Terminus 2· Terminal-Bench24%
172Qwen3 Coderterminal-bench-2.0, Terminus 2· Terminal-Bench23.9%
173Claude Opus 4.8terminal-bench-4.0, Claude Code· Terminal-Bench23.6%
174Claude Opus 4.8terminal-bench-4.1, Claude Code· Terminal-Bench23.6%
175Grok 4terminal-bench-2.0, Terminus 2· Terminal-Bench23.1%
176Kimi K3terminal-bench-4.1, Claude Code· Terminal-Bench23%
177Kimi K3terminal-bench-4.0, Claude Code· Terminal-Bench22.4%
178GPT-5 miniterminal-bench-2.0, Mini-SWE-Agent· Terminal-Bench22.2%
179Ember-1terminal-bench-4.0, mini-swe-agent· Terminal-Bench21.8%
180Ember-1terminal-bench-4.1, mini-swe-agent· Terminal-Bench21.8%
181GPT-5 nanoterminal-bench-2.0, spoox-o-m· Terminal-Bench21.8%
182GPT-5.6 Terraterminal-bench-4.0, Codex· Terminal-Bench21.5%
183GPT-5.6 Terraterminal-bench-4.1, Codex· Terminal-Bench21.5%
184Claude Opus 4.8terminal-bench-3.0, Claude Code· Terminal-Bench21.1%
185GPT-5.6 Terraterminal-bench-3.0, Codex· Terminal-Bench20.8%
186Grok 4.6terminal-bench-4.0, Grok Build· Terminal-Bench20.3%
187Grok 4.6terminal-bench-4.1, Grok Build· Terminal-Bench19.7%
188Gemini 2.5 Proterminal-bench-2.0, Gemini CLI· Terminal-Bench19.6%
189Hy4 previewterminal-bench-4.0, Claude Code· Terminal-Bench19.4%
190Gemini 3.8 Flashterminal-bench-4.0, mini-SWE-agent· Terminal-Bench19.1%
191Hy4 previewterminal-bench-4.1, Claude Code· Terminal-Bench19.1%
192GPT OSS 120Bterminal-bench-2.0, Terminus 2· Terminal-Bench18.7%
193Gemini 3.8 Flashterminal-bench-4.1, mini-SWE-agent· Terminal-Bench18.2%
194GPT-5.6 Lunaterminal-bench-4.1, Codex· Terminal-Bench17.6%
195GPT-5.6 Lunaterminal-bench-4.0, Codex· Terminal-Bench17.3%
196Gemini 2.5 Flashterminal-bench-2.0, Mini-SWE-Agent· Terminal-Bench17.1%
197Gemini 2.5 Flashterminal-bench-2.0, Terminus 2· Terminal-Bench16.9%
198Gemini 2.5 Flashterminal-bench-2.0, OpenHands· Terminal-Bench16.4%
199Gemini 2.5 Proterminal-bench-2.0, OpenHands· Terminal-Bench16.4%
200GPT-6 Lunaterminal-bench-4.0, Codex· Terminal-Bench16.4%

Showing 151–200 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)