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
1Claude Sonnet 5.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim70.6%
2Claude Sonnet 5.5Claude Code --bare · max · no internet · safeguards with default fallback· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.70.6%
3Claude Fable 5.2· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim67.6%
4Claude Opus 5.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim66.4%
5Claude Opus 5.5Opus 5.5 System Card capability section · Claude Code --bare · xhigh · safeguards with fallback · 5 trials/task· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.66.4%
6Kimi K3.1 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim65.7%
7Claude Opus 5.5Opus 5.5 System Card capability section · Claude Code --bare · max · safeguards with fallback · 5 trials/task· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.64.8%
8Claude Opus 5.5Terminal-Bench 4.0 official leaderboard · Claude Code · max effort · 95% CI ±3.1 · 8.0B tokens · $4.7k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs64.8%
9Claude Sonnet 5.5Terminal-Bench 4.0 official leaderboard · Claude Code · max effort · 95% CI ±2.9 · 19.4B tokens · $7.3k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs61.8%
10GPT-6 Astraadaptive 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.58.2%
11GPT-6 AstraTerminal-Bench 4.0 official leaderboard · Codex · max effort · 95% CI ±2.8 · 1.5B tokens · $3.3k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs58.2%
12GPT-6.1 Soladaptive 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.58.2%
13GPT-6.1 SolTerminal-Bench 4.0 official leaderboard · Codex · max effort · 95% CI ±3.1 · 1.5B tokens · $634.24· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs58.2%
14Claude Fable 5.1Harbor · 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.57.9%
15Claude Fable 5.1Terminal-Bench 4.0 official leaderboard · Claude Code · max effort · 95% CI ±3.8 · 2.7B tokens · $6.2k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs57.9%
16Claude Fable 5.1Terminal-Bench 4.0 official leaderboard · Claude Code · xhigh effort · 95% CI ±3.4 · 2.3B tokens · $4.9k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs57.9%
17GPT-6 AstraCode · high effort · chart displays 57.9% (raw 57.8788%)· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.57.9%
18GPT-6 AstraOpenAI launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.57.9%
19GPT-6 AstraOpus 5.5 System Card capability section · Codex · high effort · OpenAI-reported· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.57.9%
20GPT-6 AstraTerminal-Bench 4.0 official leaderboard · Codex · high effort · 95% CI ±3 · 1.2B tokens · $2.3k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs57.9%
21GPT-6 AstraTerminal-Bench 4.0 official leaderboard · Codex · xhigh effort · 95% CI ±2.7 · 1.2B tokens · $2.4k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs57.9%
22GPT-6 Astra· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim57.7%
23GPT-6 AstraCode · x-high effort· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.57.6%
24Gemini 4 Argon· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim57.4%
25GPT-6 AstraCode · max effort· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.56.7%
26Claude Fable 5.1System Card capability summary · adaptive thinking · max effort · default sampling · 5 trials · context ≤1M unless noted · Fable· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.56%
27Claude Fable 5.1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim55.8%
28Claude Fable 5.1Fable · Anthropic launch setting· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.55.8%
29Claude Fable 5.1Non-code · max effort· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.55.8%
30Claude Fable 5.1OpenAI launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.55.8%
31Claude Fable 5.1Opus 5.5 System Card capability section · Claude Code --bare · max· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.55.8%
32Claude Fable 5.1Terminal-Bench 4.0 official leaderboard · Claude Code · high effort · 95% CI ±3.4 · 2.2B tokens · $4.0k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs54.5%
33GPT-6 AstraTerminal-Bench 4.0 official leaderboard · Codex · medium effort · 95% CI ±2.7 · 1.1B tokens · $1.9k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs54.2%
34Claude Fable 5.1Terminal-Bench 4.0 official leaderboard · Claude Code · medium effort · 95% CI ±3.4 · 1.6B tokens · $2.8k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs53.9%
35Claude Opus 5Terminal-Bench 4.0 official leaderboard · Claude Code · xhigh effort · 95% CI ±3.2 · 6.9B tokens · $6.1k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs53.9%
36GPT-6 AstraCode · medium effort· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.53.9%
37Claude Opus 5Non-code · max effort· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.52.6%
38Claude Opus 5OpenAI launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.52.6%
39Claude Opus 5Anthropic Fable/Mythos 5.1 launch table· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.52.3%
40Claude Opus 5Opus 5.5 System Card capability section · Claude Code --bare · max· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.52.3%
41Claude Opus 5System Card capability summary · adaptive thinking · max effort · default sampling · 5 trials · context ≤1M unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.52%
42Muse Wonder 1.0· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim52%
43Claude Opus 5· LLM Stats data51.8%
44Claude Opus 5DeepSeek Harness / max effort· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.51.8%
45Claude Opus 5DeepSeek 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.51.8%
46Claude Opus 5Google Model Card· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.51.8%
47Claude Opus 5Terminal-Bench 4.0 official leaderboard · Claude Code · max effort · 95% CI ±3.4 · 6.5B tokens · $6.0k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs51.8%
48Claude Fable 5.1Non-code · x-high effort· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.51.3%
49GPT-6 AstraTerminal-Bench 4.0 official leaderboard · Codex · low effort · 95% CI ±2.8 · 889.8M tokens · $1.6k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs50.6%
50Claude Opus 5Terminal-Bench 4.0 official leaderboard · Claude Code · high effort · 95% CI ±3.7 · 5.5B tokens · $4.7k· Reported by Terminal-Bench / Stanford / Harbor / Laude Institute· via BenchBoard docs50.3%

Showing 1–50 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