AURORA Multilingual Terminal-Bench (Arabic)
Agents · 2026-09-30
The 30 Arabic-language tasks of LILT's 324-task AURORA Multilingual Terminal-Bench, with instructions authored by native speakers and run in Harbor with the Terminus-2 agent. The score is the trial-level pass rate pooled over a model's attempts on these tasks.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 75.3 |
| GPT-5.6 Sol | 74.0 |
| Claude Opus 5 | 73.3 |
| GPT-5.5 | 68.5 |
| GPT-6 Sol | 67.3 |
| Muse Spark 1.3 | 64.5 |
| Gemini 3.8 Flash | 62.8 |
| Gemini 3.5 Flash | 58.1 |
| Opus 4.8 | 46.8 |
| Mistral Medium 3.5 | 20.0 |