Skip to content

Benchmark

Terminal-Bench 2.1

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

Models and systems measured
64
Scores recorded
159
Sources
3
IQ weight
2.76

Score distribution

75.3%
Mean
82.7%
Median
17.2%
Std dev
56.4%
10th percentile
88.8%
90th percentile
2.9%
Min
2.9%91.9%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.76 for chat models, measured over 53 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 5.5Gemini 3.5 Flash Model Card · May 2026 · Terminus-2· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.78.2%
102GPT 5.5Opus 4.8 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.78.2%
103Gemini 3.6 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim78%
104Gemini 3.6 FlashGemini 3.6 Flash Model Card · July 2026 · Terminus-2· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.78%
105Muse Spark Open· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim76.4%
106Muse Spark Open 816B A42B· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim76.4%
107Gemini 3.5 FlashGemini 3.5 Flash Model Card · May 2026 · Terminus-2· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.76.2%
108Gemini 3.5 FlashGemini 3.6 Flash Model Card · July 2026 · Terminus-2· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.76.2%
109Muse Spark 1.1mini-swe-agent· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.76.2%
110Claude Opus 4.8Opus 4.8 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.74.6%
111Qwen3.7 MaxQwen表:Qwen 采用 Claude Code avg@10、5h、max_tokens 131072;友商取公开最高分· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.74.5%
112Gemini 3.1 Pro PreviewGemini 3.6 Flash Model Card · July 2026 · Terminus-2· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.73.8%
113Qwen 3.8 27B· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
73%
114Qwen 3.8 27BQwen3.8-27B model card · published setting· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.73%
115Hy3 (Deep Infra)· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
71.7%
116Hy3 (Deep Infra)Hy3 official Benchmark Appendix · highest reasoning tier unless noted · Terminus-2 · XML parser · 4-hour timeout · 500 episodes· Reported by Tencent Hunyuan· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.71.7%
117Inkling 2 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim71.6%
118Doubao Seed 2.1 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim71%
119LongCat 2.0· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim70.8%
120Gemini 3.1 Pro PreviewGemini CLI· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.70.7%
121Gemini 3.1 Pro PreviewOpenAI GPT-5.6 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.70.7%
122Gemini 3.1 Pro PreviewGemini 3.5 Flash Model Card · May 2026 · Terminus-2· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.70.3%
123Gemini 3.1 Pro Previewhigh · 89 tasks · bash-only · pass@1 · five-attempt mean· Reported by Meta, Meta Superintelligence Labs· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.70.3%
124Gemini 3.1 Pro PreviewOpus 4.8 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.70.3%
125Laguna S 2.1· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
70.2%
126Doubao Seed 2.1 Turbo· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim67.6%
127Claude Sonnet 4.6mini-SWE-agent · high effort · Sonnet 5 card infrastructure· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.67%
128NVIDIA Nemotron 3.5 Super VL 120B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
66.7%
129Claude Opus 4.7Gemini 3.5 Flash Model Card · May 2026 · Terminus-2· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.66.1%
130Claude Opus 4.7Opus 4.8 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.66.1%
131MiniMax M3· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim66%
132MiniMax M3official model-page overview figure· Reported by MiniMax· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.66%
133MiniMax M3Terminus 2 · 8C16G · 2h timeout · 128K max output· Reported by MiniMax Research· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.66%
134Inkling Small· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
64.7%
135Inkling FP4· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
63.8%
136MAI-Code-1.1-Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim62.9%
137Step 3.7 Flash· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
59.5%
138Step 3.7 FlashStep 3.7 Flash official README/figure· Reported by StepFun· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.59.5%
139Gemini 3 FlashGemini 3.5 Flash Model Card · May 2026 · Terminus-2· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.58%
140Ling 3.0 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim57%
141Solar Pro 4· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim57%
142Nemotron 3 Ultra· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
56.4%
143Nemotron 3 UltraNeMo Evaluator SDK · official table· Reported by NVIDIA· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.56.4%
144Gemini 3.5 Flash Lite· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim54%
145Muse Glimmer 30B· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
51.7%
146Muse Glimmer 30BTerminus 2 · high reasoning· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.51.7%
147Nemotron 3.5 Lightning 30B A3Bagentic-coding figure · Mini-SWE-Agent v2.4.5 · unmodified system prompt · standard harness configuration· Reported by NVIDIA· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.29.7%
148Granite 4.2 30B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim29.2%
149Nemotron 3.5 Lightning 30B A3Bagentic-coding figure · OpenCode v1.17.8 · unmodified system prompt · standard harness configuration· Reported by NVIDIA· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.29.1%
150Nemotron 3.5 Lightning 30B A3Bagentic-coding figure · Claude v2.1.126 · unmodified system prompt · standard harness configuration· Reported by NVIDIA· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.29%

Showing 101–150 of 158 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
Muse Spark (version unspecified)Meta Model API · highest effort · version unspecified · 89 tasks · bash-only · pass@1 · five-attempt mean· Reported by Meta, Meta Superintelligence Labs· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.67.3%

Data credits: Data by LLM Stats · Hugging Face official benchmark leaderboards; licence varies by benchmark · 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.