FrontierCUA
A benchmark for computer-use agents completing connected workflows across enterprise applications.
View leaderboardBuilt by Collinear AI.
Rankings
15 workflows. Each workflow counts once. A pass is a grade of at least 0.95.
| Rank | Model | Provider | Reasoning | pass@1 | avg@1 | Mean cost per rollout |
|---|---|---|---|---|---|---|
| 1 | Sol 6.1 | OpenAI | high | 46.7% | 0.629 | $5.11 |
| 2 | Astra | OpenAI | high | 42.2% | 0.575 | $41.62 |
| 3 | Opus 5 | Anthropic | high | 40.0% | 0.645 | $12.18 |
| 4 | Fable 5 | Anthropic | high | 40.0% | 0.623 | $32.87 |
| 5 | Fable 5.1 | Anthropic | high | 36.7% | 0.660 | $15.10 |
| 6 | Opus 5.5 | Anthropic | high | 36.0% | 0.686 | $7.56 |
| 7 | Sol 6 | OpenAI | high | 20.0% | 0.480 | $7.25 |
| 8 | Sonnet 5 | Anthropic | high | 20.0% | 0.431 | $10.16 |
| 9 | Kimi K3 | Moonshot | high | 20.0% | 0.373 | $46.42 |
| 10 | Qwen 3.8 Max Prime | Alibaba | high | 20.0% | 0.349 | $16.25 |
| 11 | Grok 4.7 | xAI | high | 20.0% | 0.280 | $55.63 |
| 12 | Sol 5.6 | OpenAI | high | 16.4% | 0.417 | $12.88 |
| 13 | DeepSeek V4.1 Flash | DeepSeek | high | 6.7% | 0.309 | $1.16 |
| 14 | GLM 5.3 FlashX | Z.ai | high | 6.7% | 0.291 | $7.38 |
| 15 | MiMo V2.6 Pro | Xiaomi | high | 0.0% | 0.212 | $3.19 |
Every configuration uses a computer-use adapter. DeepSeek, MiMo, Qwen, GLM, Kimi, and Grok share one screenshot and tool loop.