Screen-level OS control#13 of 15 in category
OpenAI Computer Use Preview (CUA)
OpenAI · Launched Jan 2025
OpenAI's Operator/CUA specialized model as benchmarked on OSWorld-Verified at 31.3% (max 50 steps). Now folded into ChatGPT Agent.
ChatGPT Pro
1
Benchmarks scored
31.3
Peak score
0
Article mentions
No
Open source
Benchmark performance
OSWorld-Verified
Real desktop workflows across browser, files, office apps. 369 tasks (361 without Google Drive). Human expert baseline 72.4%. Current SOTA: Holo3-35B-A3B (H Company) at 80.4%.
31.3
Other screen-level os control agents
The 15 agents in this category, ranked by peak benchmark.
| Agent | Maker | Launch | Peak | Pricing |
|---|---|---|---|---|
| Claude Sonnet 4.6 | Anthropic | 2026-02 | 1470.0 | API: 3/15 per M tokens |
| Kimi K2.5OSS | Moonshot AI | 2026-01 | 1410.0 | API pay-as-you-go |
| Claude Computer Use | Anthropic | 2024-10 | 92.1 | Claude API — input $5/M, output $25/M |
| Kimi K2.6OSS | Moonshot AI | 2026-04 | 89.6 | API: 0.60/2.75 per M tokens |
| Holo3-35B-A3B | H Company | 2026-04 | 80.4 | H Company enterprise |
| Claude Sonnet 4.5 | Anthropic | 2025-09 | 62.9 | Legacy Anthropic API |
| Seed-1.8 | ByteDance Seed | 2025-12 | 61.9 | Doubao ecosystem |
| EvoCUA-20260105 | Meituan LongCat | 2026-01 | 56.7 | Research |
| GUI-Owl-1.5 32BOSS | Alibaba Tongyi Lab | 2026-03 | 55.4 | Free (OSS) |
| DeepMiner-Mano-72B | Mininglamp Technology | 2025-10 | 53.9 | Research |
Quick facts
- Type
- Screen-level OS control
- Maker
- OpenAI
- Launch
- 2025-01-23
- Open source
- No
- Pricing
- ChatGPT Pro
- Benchmarks scored
- 1
- Article mentions
- 0
- Rank in category
- #13 of 15