FrontierCUA

FrontierCUA v1

A benchmark for computer-use agents completing connected workflows across enterprise applications.

View leaderboard

Built by Collinear AI.

Rankings

FrontierCUA leaderboard, ranked by pass@1
RankModelProviderReasoningpass@1avg@1Mean cost per rollout
1Gemini 4 CarbonGooglehigh42.5%0.606$54.47
2GPT-6.1 SolOpenAIhigh25.0%0.495$6.00
3Fable 5.2Anthropichigh25.0%0.397$49.48
4GPT-6 AstraOpenAIhigh20.0%0.440$47.94
5MiniMax M3.1 FlashMiniMaxhigh17.5%0.439$1.10
6Opus 5Anthropichigh17.5%0.431$12.30
7Fable 5.1Anthropichigh15.0%0.453$18.92
8Fable 5Anthropichigh15.0%0.440$29.10
9Opus 5.5Anthropichigh12.5%0.442$10.52
10Muse WonderMetahigh12.5%0.419$55.08
11Kimi K3Moonshothigh12.5%0.392$8.23
12Sonnet 5Anthropichigh12.5%0.381$8.76
13MiMo V2.6 FlashXiaomihigh12.5%0.322$1.31
14Muse EmberMetahigh7.5%0.408$1.37
15GPT-6 SolOpenAIhigh7.5%0.307$5.18
16GPT-6.1 TerraOpenAIhigh7.5%0.297$0.92
17Grok 4.7xAIhigh7.5%0.275$71.96
18GPT-5.6 SolOpenAIhigh7.5%0.271$7.43
19MiMo V2.6 ProXiaomihigh5.0%0.276$2.48
20Kimi K3.1 ProMoonshothigh2.5%0.282$4.91
21GLM 5.3 FlashXZ.aihigh2.5%0.257$5.94
22DeepSeek V4.1 FlashDeepSeekhigh2.5%0.243$1.12
23Qwen 3.8 Max PrimeAlibabahigh0.0%0.201$28.15
24Muse Spark 1.3Metahigh0.0%0.110$54.32

Every configuration uses a computer-use adapter.