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
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 | |
|---|---|---|
| 51 | GPT 5.2terminal-bench-2.0, Codex CLI· Terminal-Bench | 62.9% |
| 52 | Gemini 3 Pro Previewterminal-bench-2.0, CodeBrain-1.5· Terminal-Bench | 62.2% |
| 53 | GPT-6.2 Astraterminal-bench-4.1, Codex· Terminal-Bench | 62.1% |
| 54 | Gemini 3 Pro Previewterminal-bench-2.0, II-Agent· Terminal-Bench | 61.8% |
| 55 | GPT-5.1 Codex Miniterminal-bench-2.0, hookele· Terminal-Bench | 61.6% |
| 56 | GPT-6.2 Solterminal-bench-4.1, Codex· Terminal-Bench | 61.5% |
| 57 | Gemini 3.1 Pro Previewterminal-bench-2.0, Gemini CLI· Terminal-Bench | 61.4% |
| 58 | Gemini 4 Neonterminal-bench-4.1, Antigravity CLI· Terminal-Bench | 61.2% |
| 59 | Gemini 3 Pro Previewterminal-bench-2.0, Droid· Terminal-Bench | 61.1% |
| 60 | GPT-6.1 Solterminal-bench-4.1, Codex· Terminal-Bench | 60.9% |
| 61 | Kimi K3.1 Proterminal-bench-4.0, Claude Code· Terminal-Bench | 60.9% |
| 62 | GPT 5.2terminal-bench-2.0, Mux· Terminal-Bench | 60.7% |
| 63 | GPT-5.1 Codex Maxterminal-bench-2.0, Codex CLI· Terminal-Bench | 60.4% |
| 64 | GPT-6 Astraterminal-bench-4.1, Codex· Terminal-Bench | 60.3% |
| 65 | Kimi K3.1 Proterminal-bench-4.1, Claude Code· Terminal-Bench | 59.7% |
| 66 | Claude Opus 4.5terminal-bench-2.0, Letta Code· Terminal-Bench | 59.1% |
| 67 | GLM 5.1terminal-bench-2.1, Claude Code· Terminal-Bench | 58.7% |
| 68 | Claude Opus 4.5terminal-bench-2.0, Mux· Terminal-Bench | 58.4% |
| 69 | GPT-6 Astraterminal-bench-4.0, Codex· Terminal-Bench | 58.2% |
| 70 | GPT-6.1 Solterminal-bench-4.0, Codex· Terminal-Bench | 58.2% |
| 71 | Claude Opus 4.6terminal-bench-2.0, Claude Code· Terminal-Bench | 58% |
| 72 | Claude Opus 4.5terminal-bench-2.0, Terminus 2· Terminal-Bench | 57.8% |
| 73 | GPT-5.1 Codexterminal-bench-2.0, Crux· Terminal-Bench | 57.8% |
| 74 | Grok 4.20terminal-bench-2.0, Grok CLI· Terminal-Bench | 57.3% |
| 75 | Gemini 3 Pro Previewterminal-bench-2.0, Terminus 2· Terminal-Bench | 56.9% |
| 76 | Gemini 4 Hydrogenterminal-bench-4.1, Antigravity CLI· Terminal-Bench | 56.4% |
| 77 | Gemini 3 Pro Previewterminal-bench-2.0, Letta Code· Terminal-Bench | 56% |
| 78 | Gemini 4 Argonterminal-bench-4.1, Antigravity CLI· Terminal-Bench | 55.5% |
| 79 | Gemini 4 Argonterminal-bench-4.0, Antigravity CLI· Terminal-Bench | 54.9% |
| 80 | Claude Opus 4.5terminal-bench-2.0, Goose· Terminal-Bench | 54.3% |
| 81 | GPT 5.2terminal-bench-2.0, Terminus 2· Terminal-Bench | 54% |
| 82 | GPT-5.1 Codexterminal-bench-2.0, Letta Code· Terminal-Bench | 53.5% |
| 83 | Claude Sonnet 4.6terminal-bench-2.0, Simplai Agent· Terminal-Bench | 53.4% |
| 84 | GPT-6.1 Terraterminal-bench-4.1, Codex· Terminal-Bench | 53.3% |
| 85 | GLM 5terminal-bench-2.0, Terminus 2· Terminal-Bench | 52.4% |
| 86 | Claude Opus 4.5terminal-bench-2.0, Claude Code· Terminal-Bench | 52.1% |
| 87 | Claude Opus 4.5terminal-bench-2.0, OpenHands· Terminal-Bench | 51.9% |
| 88 | GPT-6.1 Terraterminal-bench-4.0, Codex· Terminal-Bench | 51.8% |
| 89 | Claude Opus 4.5terminal-bench-2.0, OpenCode· Terminal-Bench | 51.7% |
| 90 | Gemini 3 Flashterminal-bench-2.0, Terminus 2· Terminal-Bench | 51.7% |
| 91 | GPT-6 Solterminal-bench-4.1, Codex· Terminal-Bench | 50.3% |
| 92 | GPT 5terminal-bench-2.0, Codex CLI· Terminal-Bench | 49.6% |
| 93 | GPT-6 Solterminal-bench-4.0, Codex· Terminal-Bench | 49.4% |
| 94 | GPT 5.1terminal-bench-2.0, Terminus 2· Terminal-Bench | 47.6% |
| 95 | Gemini 3 Flashterminal-bench-2.0, Gemini CLI· Terminal-Bench | 47.4% |
| 96 | Muse Wonderterminal-bench-4.0, Muse Code· Terminal-Bench | 47% |
| 97 | Muse Wonderterminal-bench-4.1, Muse Code· Terminal-Bench | 47% |
| 98 | Claude Sonnet 4.5terminal-bench-2.0, CAMEL-AI· Terminal-Bench | 46.5% |
| 99 | MiniMax M2.7terminal-bench-2.0, IndusAGI Coding Agent· Terminal-Bench | 45.1% |
| 100 | GPT-5 Codexterminal-bench-2.0, Codex CLI· Terminal-Bench | 44.3% |
Showing 51–100 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)