CUA-bench
Playing six held-out commercial games through keyboard and mouse interaction.
Computer useVals AI8 models run by Vals AIUpdated 29 September 2026
How it is run
The agent observes the game and selects keyboard and mouse actions. Vals reports the overall score across its held-out environments.
What it cannot tell you
Game interaction tests a particular computer-use protocol; it is not a general desktop-automation or browser-workflow score.
- Leader
- GPT-6 Astra19.17% · Vals AI's own list
- Top-ten spread
- 15.00%Leader minus tenth place, or last place in smaller fields.
- Median
- 8.66%of the 8 rows published
- No published result
- 38of 46 models on this site
Independent leaderboard · Vals AI
Top 8 of 8 runs by Vals AI, one harness.
| # | Model | Score |
|---|---|---|
| 1 | GPT-6 Astra | 19.17% |
| 2 | Claude Opus 5.5 | 14.00% |
| 3 | Claude Fable 5.1 | 13.17% |
| 4 | Claude Opus 5 | 9.00% |
| 5 | GPT-5.6 Sol | 8.33% |
| 6 | Muse Spark 1.3 Max | 5.83% |
| 7 | Gemini 4 Argon | 4.83% |
| 8 | Gemini 3.8 Flash | 4.17% |
No published result tracked
38 of the 46 models on this site have no published number on this benchmark, so their cells in the home table stay blank rather than reading as zero.
Step 5 PreviewGPT-6 LunaGPT-5.6 TerraGLM 5.3 FlashQwen3.8 2.4T A95BQwen3.8-Flash-NextMiMo-V2.6-FlashQwen3.8 Max (0902)Claude Fable 5Grok 4.6Claude Opus 4.8Claude Opus 4.7Gemini 3.7 FlashMuse Spark 1.2GPT-5.4Grok 4.5GPT-5.5DeepSeek V4.1 Flash 1017GLM-LegalClaude Haiku 5.5BeamClaude Sonnet 5.5Claude Sonnet 5GPT-6 SolGPT-6.1 SolGPT-5.6 LunaGemini 3.1 ProGrok 4.7DeepSeek V4.1 FlashDeepSeek V4 ProKimi K3Qwen3.8 MaxGLM 5.3MiniMax M3Hunyuan Hy4 PreviewMiMo V2.6 ProInklingMistral Medium 3.5
Read this with
Benchmarks in the same category, which usually disagree with this one on purpose.
Sources
Every page used for this benchmark, read on 17 October 2026.
- Benchmark site or the page that publishes itwww.vals.ai/benchmarks/cua_bench
- Independent resultswww.vals.ai/benchmarks/cua_bench