Skip to content

Benchmark

Terminal-Bench 4.0

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

Models and systems measured
44
Scores recorded
136
Sources
2
IQ weight
1.66

Score distribution

40%
Mean
41.8%
Median
16.8%
Std dev
14.5%
10th percentile
57.9%
90th percentile
1%
Min
1%70.6%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.66 for chat models, measured over 35 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model or systemScore
101GPT-5.6 SolNon-code · high effort· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.26.1%
102MiniMax M3.1 FlashVals AI October 20 model evaluation · official MiniMax API · temperature 1 / top_p 0.95 / compute effort max · displayed uncertainty ±1.01 percentage points · output cap inconsistent across the page· Reported by Vals AI· via BenchBoard docs25.3%
103Claude Opus 4.8· LLM Stats data23.6%
104Claude Opus 4.8Terminal-Bench 4.0 official leaderboard · Claude Code · max effort · 95% CI ±3.6 · 6.4B tokens · $6.5k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs23.6%
105GPT-5.6 Terra· LLM Stats data21.5%
106GPT-5.6 TerraHarbor · provider harness · task success rate · max effort· Reported by SpaceXAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.21.5%
107GPT-5.6 TerraTerminal-Bench 4.0 official leaderboard · Codex · max effort · 95% CI ±3.3 · 5.7B tokens · $1.7k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs21.5%
108GPT-5.6 SolNon-code · medium effort· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.20.9%
109Grok 4.6· LLM Stats data20.3%
110Hy4 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim19.2%
111Gemini 3.8 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim19.1%
112Gemini 3.8 FlashGoogle Model Card· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.19.1%
113Gemini 3.8 FlashOpenAI launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.19.1%
114Gemini 3.8 FlashTerminal-Bench 4.0 official leaderboard · mini-SWE-agent · high effort · 95% CI ±3.4 · 17.2B tokens · $1.8k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs19.1%
115GPT-5.6 Luna· LLM Stats data17.3%
116GPT-5.6 LunaTerminal-Bench 4.0 official leaderboard · Codex · max effort · 95% CI ±2.8 · 11.6B tokens · $346.67· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs17.3%
117GPT-6 Lunaadaptive thinking · max effort · default sampling · 5-trial mean · context ≤1M unless noted · public leaderboard · Codex CLI · max· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.16.4%
118GPT-6 LunaTerminal-Bench 4.0 official leaderboard · Codex · max effort · 95% CI ±2.7 · 3.9B tokens · $72.14· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs16.4%
119Muse Spark 1.3Terminal-Bench 4.0 official leaderboard · Muse Code · xhigh effort · 95% CI ±2.9 · 9.0B tokens · $0.00 as displayed· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs14.5%
120Kimi K3DeepSeek Harness / max effort· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.12.6%
121Kimi K3DeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.12.6%
122Claude Sonnet 5· LLM Stats data12.4%
123Claude Sonnet 5Harbor · provider harness · task success rate · max effort· Reported by SpaceXAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.12.4%
124Claude Sonnet 5Terminal-Bench 4.0 official leaderboard · Claude Code · max effort · 95% CI ±3.1 · 21.6B tokens · $9.6k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs12.4%
125DeepSeek V4 Pro 0813DeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.12.4%
126Grok 4.5· LLM Stats data12.4%
127Grok 4.5Harbor · provider harness · task success rate · high effort· Reported by SpaceXAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.12.4%
128Grok 4.5Terminal-Bench 4.0 official leaderboard · Grok Build · high effort · 95% CI ±2.6 · 3.4B tokens · $2.1k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs12.4%
129Gemini 3.7 FlashTerminal-Bench 4.0 official leaderboard · mini-SWE-agent · high effort · 95% CI ±2.4 · 11.1B tokens · $1.3k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs11.2%
130GPT-5.6 SolNon-code · low effort· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.7.9%
131DeepSeek V4 Flash 0731DeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.7%
132Inkling 2 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim5.6%
133Solar Mini 4· LLM Stats data1%

Showing 101–133 of 133 results.

Not ranked

The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.

Model or systemScore
Grok 4.6Harbor · provider harness · task success rate · high effort· Reported by SpaceXAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.20.3%
Grok 4.6Terminal-Bench 4.0 official leaderboard · Grok Build · high effort · 95% CI ±3.1 · 4.0B tokens · $3.6k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs20.3%
Qwen3.8 Max 0902Terminal-Bench 4.0 official leaderboard · Claude Code · max effort · 95% CI ±3.8 · 3.0B tokens · $4.6k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs27%

Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · Data by LLM Stats