DeskCraft
Desktop agents complete 538 Ubuntu tasks, including 152 tasks with later user clarifications or interruptions. Programmatic checks inspect final application state, saved files, and exported artifacts.
The suite includes 386 standard tasks. Interactive tasks use explicit triggers to introduce later user messages.
- Made by
- Zhejiang University, Tsinghua University, Tencent and others
- Released
- June 2026
- Kind
- Desktop apps
- Level
- Expert
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Measures
- Capability
- Subject
- Arts, Other engineering and technologies
- Language
- English
- Interaction
- Agentic
- Environment
- Desktop apps
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Computer use, Agents
- Strongest model tested
- GPT-5.4 #28 on the capability index
- Last model added
- April 2026
- Scores refreshed
- November 20th, 2026
In the authors’ words
Real-world professional desktop workflows in specialized creative and engineering software unfold over long horizons and often require human-in-the-loop coordination, where agents proactively seek necessary information and users provide additional instructions, clarifications, feedback, or corrections as the task progresses. Yet existing desktop GUI benchmarks mostly reduce this setting to short, simplified tasks with all user instructions provided upfront. To address this issue, we introduce DeskCraft, a desktop GUI benchmark targeting long horizon creative and engineering workflows and proactive human-agent collaboration. DeskCraft organizes tasks into a multilevel difficulty taxonomy, with long horizon tasks requiring over 50 execution steps, and covers professional creative software across design, video, audio, and 3D creation. Furthermore, DeskCraft formalizes human-agent collaboration into an interaction protocol covering mid-turn and post-turn exchanges. Mid-turn interaction captures both agent-initiated clarification under uncertainty and user-initiated interruption during execution, while post-turn interaction accommodates user-driven feedback after the agent signals completion, together spanning the full space of realistic collaboration patterns. We evaluate 18 proprietary and open source agents on 538 tasks and find that GPT-5.4 reaches 31.6% on standard tasks and 27.6% on interactive tasks. Further analyses reveal persistent failures in long horizon workflow delivery and proactive clarification. We will open-source all evaluation codes, tasks, and data at https://github.com/mrwwk/DeskCraft.
Kimi K2.6 leads with 29.8% among models released by April 2026 Kimi K2.6 leads with 29.8% among models released by April 2026
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper reports success separately for Standard and Interactive tasks, so the headline is the plain mean of the two rates. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in April 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 29.8 Kimi K2.6
- 29.6 GPT-5.4
- 22.1 Kimi K2.5
- 12.3
Qwen3.5 397B-A17B
- 11.8
Qwen3.5-35B-A3B
- 8.55
Qwen 3.6 35B-A3B
- 6.85 Qwen3-VL-32B
- 6.45
Qwen3.5-9B
- 5.70 EvoCUA-32B
- 5.25
Qwen3-VL-235B-A22B
- 4.80 OpenCUA-32B
- 3.85
EvoCUA-8B
- 2.95 OpenCUA-7B
- 2.70 GUI-Owl-1.5-8B-Instruct
- 2.65 Qwen3-VL-8B
- 2.40 GUI-Owl-1.5-32B-Instruct
- 1.55
UI-TARS-1.5
- 0.15 OS-Atlas-Pro-7B
- 1 Kimi K2.6 29.8%
- 2 GPT-5.4 29.6%
- 3 Kimi K2.5 22.1%
- 4
Qwen3.5 397B-A17B 12.3%
- 5
Qwen3.5-35B-A3B 11.8%
- 6
Qwen 3.6 35B-A3B 8.55%
- 7 Qwen3-VL-32B 6.85%
- 8
Qwen3.5-9B 6.45%
- 9 EvoCUA-32B 5.70%
- 10
Qwen3-VL-235B-A22B 5.25%
- 11 OpenCUA-32B 4.80%
- 12
EvoCUA-8B 3.85%
- 13 OpenCUA-7B 2.95%
- 14 GUI-Owl-1.5-8B-Instruct 2.70%
- 15 Qwen3-VL-8B 2.65%
- 16 GUI-Owl-1.5-32B-Instruct 2.40%
- 17
UI-TARS-1.5 1.55%
- 18 OS-Atlas-Pro-7B 0.15%
Source: The DeskCraft paper
- 1 Kimi K2.6 29.8%
- 2 GPT-5.4 29.6%
- 3 Kimi K2.5 22.1%
- 4
Qwen3.5 397B-A17B 12.3%
- 5
Qwen3.5-35B-A3B 11.8%
- 6
Qwen 3.6 35B-A3B 8.55%
- 7
Qwen3.5-9B 6.45%
- 8
Qwen3-VL-235B-A22B 5.25%
- 9 OpenCUA-32B 4.80%
- 10
EvoCUA-8B 3.85%
- 11 OpenCUA-7B 2.95%
- 12
UI-TARS-1.5 1.55%
Source: The DeskCraft paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Kimi K2.6 | 29.8% | Apr 2026 | Official leaderboard | |
| 2 | GPT-5.4 | 29.6% | Mar 2026 | Official leaderboard | |
| 3 | Kimi K2.5 | 22.1% | Jan 2026 | Official leaderboard | |
| 4 | | 12.3% | Feb 2026 | Official leaderboard | |
| 5 | | 11.8% | Feb 2026 | Official leaderboard | |
| 6 | | 8.55% | Apr 2026 | Official leaderboard | |
| 7 | Qwen3-VL-32B | 6.85% | Official leaderboard | ||
| 8 | | 6.45% | Feb 2026 | Official leaderboard | |
| 9 | EvoCUA-32B | 5.70% | Official leaderboard | ||
| 10 | | 5.25% | Sep 2025 | Official leaderboard | |
| 11 | OpenCUA-32B | 4.80% | Jul 2025 | Official leaderboard | |
| 12 | | 3.85% | Jan 2026 | Official leaderboard | |
| 13 | OpenCUA-7B | 2.95% | Jul 2025 | Official leaderboard | |
| 14 | GUI-Owl-1.5-8B-Instruct | 2.70% | Official leaderboard | ||
| 15 | Qwen3-VL-8B | 2.65% | Official leaderboard | ||
| 16 | GUI-Owl-1.5-32B-Instruct | 2.40% | Official leaderboard | ||
| 17 | | 1.55% | Apr 2025 | Official leaderboard | |
| 18 | OS-Atlas-Pro-7B | 0.15% | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Kimi K2.6 | 29.8% | Apr 2026 | Official leaderboard | |
| 2 | GPT-5.4 | 29.6% | Mar 2026 | Official leaderboard | |
| 3 | Kimi K2.5 | 22.1% | Jan 2026 | Official leaderboard | |
| 4 | | 12.3% | Feb 2026 | Official leaderboard | |
| 5 | | 11.8% | Feb 2026 | Official leaderboard | |
| 6 | | 8.55% | Apr 2026 | Official leaderboard | |
| 7 | | 6.45% | Feb 2026 | Official leaderboard | |
| 8 | | 5.25% | Sep 2025 | Official leaderboard | |
| 9 | OpenCUA-32B | 4.80% | Jul 2025 | Official leaderboard | |
| 10 | | 3.85% | Jan 2026 | Official leaderboard | |
| 11 | OpenCUA-7B | 2.95% | Jul 2025 | Official leaderboard | |
| 12 | | 1.55% | Apr 2025 | Official leaderboard |
Rankings by part
We read 28 parts from the benchmark’s paper in 2 groups: standard and interactive. Each ranks models on one part of the test. Kimi K2.6 leads 9 of them, and 6 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Standard
12 models we track have a score here.
- 1 Kimi K2.6 33.8%
- 2 GPT-5.4 31.6%
- 3 Kimi K2.5 20.3%
- 4
Qwen3.5 397B-A17B 12.9%
- 5
Qwen3.5-35B-A3B 11.1%
- 6
Qwen 3.6 35B-A3B 10.2%
- 7 OpenCUA-32B ↑2 9.60%
- 8
Qwen3.5-9B ↓1 8.70%
- 9
EvoCUA-8B ↑1 7.00%
- 10 OpenCUA-7B ↑1 5.20%
Interactive
12 models we track have a score here.
- 1 GPT-5.4 ↑1 27.6%
- 2 Kimi K2.6 ↓1 25.7%
- 3 Kimi K2.5 24%
- 4
Qwen3.5-35B-A3B ↑1 12.4%
- 5
Qwen3.5 397B-A17B ↓1 11.7%
- 6
Qwen 3.6 35B-A3B 6.90%
- 7
Qwen3-VL-235B-A22B ↑1 6.20%
- 8
Qwen3.5-9B ↓1 4.20%
- 9 OpenCUA-7B ↑2 0.70%
- 10
EvoCUA-8B 0.70%
Writer
12 models we track have a score here.
- 1 GPT-5.4 ↑1 50%
- 2 Kimi K2.5 ↑1 33.3%
- 3
Qwen3-VL-235B-A22B ↑5 8.30%
- 4
EvoCUA-8B ↑6 8.30%
- 5
Qwen3.5 397B-A17B ↓1 8.30%
- 6
Qwen3.5-35B-A3B ↓1 8.30%
- 7
Qwen3.5-9B 8.30%
- 8
Qwen 3.6 35B-A3B ↓2 8.30%
- 9 Kimi K2.6 ↓8 8.30%
- 10
UI-TARS-1.5 ↑2 0%
Calc
12 models we track have a score here.
- 1 Kimi K2.6 27.8%
- 2 GPT-5.4 22.2%
- 3
Qwen3-VL-235B-A22B ↑5 11.1%
- 4 Kimi K2.5 ↓1 11.1%
- 5
Qwen3.5-9B ↑2 11.1%
- 6
EvoCUA-8B ↑4 5.60%
- 7
Qwen3.5 397B-A17B ↓3 5.60%
- 8
UI-TARS-1.5 ↑4 0%
- 9 OpenCUA-32B 0%
- 10 OpenCUA-7B ↑1 0%
Impress
12 models we track have a score here.
- 1
Qwen3.5-35B-A3B ↑4 6.90%
- 2 GPT-5.4 6.90%
- 3
Qwen 3.6 35B-A3B ↑3 6.90%
- 4 Kimi K2.6 ↓3 3.70%
- 5
UI-TARS-1.5 ↑7 3.40%
- 6 OpenCUA-32B ↑3 3.40%
- 7 OpenCUA-7B ↑4 3.40%
- 8
EvoCUA-8B ↑2 3.40%
- 9 Kimi K2.5 ↓6 3.40%
- 10
Qwen3.5 397B-A17B ↓6 3.40%
Chrome
12 models we track have a score here.
- 1 GPT-5.4 ↑1 30.3%
- 2 Kimi K2.6 ↓1 28%
- 3
Qwen3.5 397B-A17B ↑1 20%
- 4
Qwen3.5-9B ↑3 20%
- 5 OpenCUA-7B ↑6 18.2%
- 6
EvoCUA-8B ↑4 15.2%
- 7
Qwen3.5-35B-A3B ↓2 13%
- 8
Qwen3-VL-235B-A22B 12%
- 9
Qwen 3.6 35B-A3B ↓3 12%
- 10 OpenCUA-32B ↓1 9.10%
VS Code
12 models we track have a score here.
- 1 GPT-5.4 ↑1 50%
- 2 Kimi K2.6 ↓1 37.9%
- 3 Kimi K2.5 20%
- 4 OpenCUA-32B ↑5 16.7%
- 5
Qwen3.5-9B ↑2 14.3%
- 6 OpenCUA-7B ↑5 13.3%
- 7
EvoCUA-8B ↑3 13.3%
- 8
Qwen3.5 397B-A17B ↓4 10.3%
- 9
UI-TARS-1.5 ↑3 10%
- 10
Qwen3.5-35B-A3B ↓5 7.10%
GIMP
12 models we track have a score here.
- 1 GPT-5.4 ↑1 17.4%
- 2 Kimi K2.5 ↑1 13%
- 3 Kimi K2.6 ↓2 13%
- 4 OpenCUA-32B ↑5 8.70%
- 5
Qwen3.5-35B-A3B 8.70%
- 6
Qwen3-VL-235B-A22B ↑2 4.30%
- 7
Qwen3.5 397B-A17B ↓3 4.30%
- 8
Qwen3.5-9B ↓1 4.30%
- 9
UI-TARS-1.5 ↑3 0%
- 10 OpenCUA-7B ↑1 0%
Inkscape
12 models we track have a score here.
- 1 Kimi K2.6 65.4%
- 2 GPT-5.4 50%
- 3 Kimi K2.5 46.2%
- 4
Qwen 3.6 35B-A3B ↑2 19.2%
- 5 OpenCUA-32B ↑4 11.5%
- 6
Qwen3.5 397B-A17B ↓2 11.5%
- 7
Qwen3.5-35B-A3B ↓2 11.5%
- 8
Qwen3.5-9B ↓1 11.5%
- 9
EvoCUA-8B ↑1 7.70%
- 10
UI-TARS-1.5 ↑2 3.80%
Kdenlive
12 models we track have a score here.
- 1 GPT-5.4 ↑1 27.9%
- 2 Kimi K2.6 ↓1 27.9%
- 3 Kimi K2.5 25.6%
- 4
Qwen3.5-35B-A3B ↑1 18.6%
- 5
Qwen 3.6 35B-A3B ↑1 18.6%
- 6
Qwen3.5 397B-A17B ↓2 16.3%
- 7 OpenCUA-32B ↑2 14%
- 8
Qwen3.5-9B ↓1 7.10%
- 9 OpenCUA-7B ↑2 4.70%
- 10
UI-TARS-1.5 ↑2 2.30%
Audacity
12 models we track have a score here.
- 1 Kimi K2.6 48.4%
- 2 GPT-5.4 32.3%
- 3 Kimi K2.5 25.8%
- 4 OpenCUA-32B ↑5 12.9%
- 5
Qwen3.5-35B-A3B 12.9%
- 6
Qwen3.5 397B-A17B ↓2 9.70%
- 7 OpenCUA-7B ↑4 6.50%
- 8
Qwen3.5-9B ↓1 6.50%
- 9
Qwen 3.6 35B-A3B ↓3 3.20%
- 10
UI-TARS-1.5 ↑2 0%
Blender
12 models we track have a score here.
- 1 Kimi K2.6 67.6%
- 2 GPT-5.4 54.1%
- 3 Kimi K2.5 43.2%
- 4
Qwen3.5 397B-A17B 18.9%
- 5
Qwen 3.6 35B-A3B ↑1 18.9%
- 6
Qwen3.5-35B-A3B ↓1 14.3%
- 7 OpenCUA-32B ↑2 13.5%
- 8
Qwen3-VL-235B-A22B 8.10%
- 9
Qwen3.5-9B ↓2 8.10%
- 10
UI-TARS-1.5 ↑2 5.40%
UI Gen
12 models we track have a score here.
- 1 Kimi K2.6 18.5%
- 2
EvoCUA-8B ↑8 10%
- 3 GPT-5.4 ↓1 10%
- 4 Kimi K2.5 ↓1 6.70%
- 5
Qwen3.5 397B-A17B ↓1 4.50%
- 6
Qwen3.5-9B ↑1 3.70%
- 7
UI-TARS-1.5 ↑5 0%
- 8 OpenCUA-32B ↑1 0%
- 9 OpenCUA-7B ↑2 0%
- 10
Qwen3-VL-235B-A22B ↓2 0%
Multi-app
12 models we track have a score here.
- 1 GPT-5.4 ↑1 30.2%
- 2 Kimi K2.6 ↓1 25.6%
- 3
Qwen3.5-35B-A3B ↑2 17.9%
- 4
Qwen3.5 397B-A17B 15%
- 5
Qwen 3.6 35B-A3B ↑1 10%
- 6 OpenCUA-32B ↑3 9.30%
- 7
EvoCUA-8B ↑3 9.30%
- 8 Kimi K2.5 ↓5 9.30%
- 9
Qwen3-VL-235B-A22B ↓1 7.50%
- 10
Qwen3.5-9B ↓3 5.00%
OS
12 models we track have a score here.
- 1 Kimi K2.6 41.4%
- 2 GPT-5.4 32.3%
- 3 Kimi K2.5 29%
- 4
Qwen3.5 397B-A17B 26.7%
- 5
Qwen 3.6 35B-A3B ↑1 16.7%
- 6
Qwen3.5-9B ↑1 13.8%
- 7 OpenCUA-32B ↑2 12.9%
- 8
Qwen3.5-35B-A3B ↓3 12.9%
- 9
EvoCUA-8B ↑1 9.70%
- 10
Qwen3-VL-235B-A22B ↓2 6.50%
Writer
12 models we track have a score here.
- 1 Kimi K2.5 ↑2 81.8%
- 2 GPT-5.4 81.8%
- 3 Kimi K2.6 ↓2 54.5%
- 4
Qwen3-VL-235B-A22B ↑4 45.5%
- 5
Qwen3.5-35B-A3B 45.5%
- 6
Qwen 3.6 35B-A3B 45.5%
- 7
Qwen3.5 397B-A17B ↓3 36.4%
- 8
Qwen3.5-9B ↓1 36.4%
- 9
UI-TARS-1.5 ↑3 0%
- 10 OpenCUA-32B ↓1 0%
Calc
12 models we track have a score here.
- 1
UI-TARS-1.5 ↑11 0%
- 2 OpenCUA-32B ↑7 0%
- 3 OpenCUA-7B ↑8 0%
- 4
Qwen3-VL-235B-A22B ↑4 0%
- 5
EvoCUA-8B ↑5 0%
- 6 Kimi K2.5 ↓3 0%
- 7
Qwen3.5 397B-A17B ↓3 0%
- 8
Qwen3.5-35B-A3B ↓3 0%
- 9
Qwen3.5-9B ↓2 0%
- 10 GPT-5.4 ↓8 0%
Impress
12 models we track have a score here.
- 1
Qwen3.5 397B-A17B ↑3 10%
- 2 GPT-5.4 10%
- 3
UI-TARS-1.5 ↑9 0%
- 4 OpenCUA-32B ↑5 0%
- 5 OpenCUA-7B ↑6 0%
- 6
Qwen3-VL-235B-A22B ↑2 0%
- 7
EvoCUA-8B ↑3 0%
- 8 Kimi K2.5 ↓5 0%
- 9
Qwen3.5-35B-A3B ↓4 0%
- 10
Qwen3.5-9B ↓3 0%
Chrome
12 models we track have a score here.
- 1 Kimi K2.5 ↑2 9.10%
- 2 Kimi K2.6 ↓1 7.10%
- 3
Qwen3-VL-235B-A22B ↑5 4.80%
- 4
Qwen3.5 397B-A17B 4.80%
- 5
Qwen3.5-35B-A3B 4.80%
- 6
UI-TARS-1.5 ↑6 0%
- 7 OpenCUA-32B ↑2 0%
- 8 OpenCUA-7B ↑3 0%
- 9
EvoCUA-8B ↑1 0%
- 10
Qwen3.5-9B ↓3 0%
VS Code
12 models we track have a score here.
- 1 GPT-5.4 ↑1 25%
- 2 Kimi K2.6 ↓1 12.5%
- 3 Kimi K2.5 4.80%
- 4
Qwen3.5 397B-A17B 4.80%
- 5
UI-TARS-1.5 ↑7 0%
- 6 OpenCUA-32B ↑3 0%
- 7 OpenCUA-7B ↑4 0%
- 8
Qwen3-VL-235B-A22B 0%
- 9
EvoCUA-8B ↑1 0%
- 10
Qwen3.5-35B-A3B ↓5 0%
GIMP
12 models we track have a score here.
- 1
UI-TARS-1.5 ↑11 0%
- 2 OpenCUA-32B ↑7 0%
- 3 OpenCUA-7B ↑8 0%
- 4
Qwen3-VL-235B-A22B ↑4 0%
- 5
EvoCUA-8B ↑5 0%
- 6 Kimi K2.5 ↓3 0%
- 7
Qwen3.5 397B-A17B ↓3 0%
- 8
Qwen3.5-35B-A3B ↓3 0%
- 9
Qwen3.5-9B ↓2 0%
- 10 GPT-5.4 ↓8 0%
Inkscape
12 models we track have a score here.
- 1 Kimi K2.6 28.6%
- 2
UI-TARS-1.5 ↑10 0%
- 3 OpenCUA-32B ↑6 0%
- 4 OpenCUA-7B ↑7 0%
- 5
Qwen3-VL-235B-A22B ↑3 0%
- 6
EvoCUA-8B ↑4 0%
- 7 Kimi K2.5 ↓4 0%
- 8
Qwen3.5 397B-A17B ↓4 0%
- 9
Qwen3.5-35B-A3B ↓4 0%
- 10
Qwen3.5-9B ↓3 0%
Kdenlive
12 models we track have a score here.
- 1
UI-TARS-1.5 ↑11 0%
- 2 OpenCUA-32B ↑7 0%
- 3 OpenCUA-7B ↑8 0%
- 4
Qwen3-VL-235B-A22B ↑4 0%
- 5
EvoCUA-8B ↑5 0%
- 6 Kimi K2.5 ↓3 0%
- 7
Qwen3.5 397B-A17B ↓3 0%
- 8
Qwen3.5-35B-A3B ↓3 0%
- 9
Qwen3.5-9B ↓2 0%
- 10 GPT-5.4 ↓8 0%
Audacity
12 models we track have a score here.
- 1
Qwen3-VL-235B-A22B ↑7 40%
- 2 Kimi K2.6 ↓1 40%
- 3 Kimi K2.5 25%
- 4
Qwen3.5 397B-A17B 20%
- 5
Qwen3.5-35B-A3B 20%
- 6
Qwen3.5-9B ↑1 20%
- 7 GPT-5.4 ↓5 20%
- 8
UI-TARS-1.5 ↑4 0%
- 9 OpenCUA-32B 0%
- 10 OpenCUA-7B ↑1 0%
Blender
12 models we track have a score here.
- 1
UI-TARS-1.5 ↑11 0%
- 2 OpenCUA-32B ↑7 0%
- 3 OpenCUA-7B ↑8 0%
- 4
Qwen3-VL-235B-A22B ↑4 0%
- 5
EvoCUA-8B ↑5 0%
- 6 Kimi K2.5 ↓3 0%
- 7
Qwen3.5 397B-A17B ↓3 0%
- 8
Qwen3.5-35B-A3B ↓3 0%
- 9
Qwen3.5-9B ↓2 0%
- 10 GPT-5.4 ↓8 0%
UI Gen
12 models we track have a score here.
- 1 Kimi K2.6 33.3%
- 2
Qwen3.5-35B-A3B ↑3 22.2%
- 3 GPT-5.4 ↓1 22.2%
- 4
Qwen3.5 397B-A17B 11.1%
- 5
Qwen3.5-9B ↑2 5.60%
- 6
Qwen 3.6 35B-A3B 5.60%
- 7
UI-TARS-1.5 ↑5 0%
- 8 OpenCUA-32B ↑1 0%
- 9 OpenCUA-7B ↑2 0%
- 10
Qwen3-VL-235B-A22B ↓2 0%
Multi-app
12 models we track have a score here.
- 1 Kimi K2.5 ↑2 52%
- 2 GPT-5.4 48%
- 3 Kimi K2.6 ↓2 44%
- 4 OpenCUA-7B ↑7 4.00%
- 5
EvoCUA-8B ↑5 4.00%
- 6
UI-TARS-1.5 ↑6 0%
- 7 OpenCUA-32B ↑2 0%
- 8
Qwen3-VL-235B-A22B 0%
- 9
Qwen3.5 397B-A17B ↓5 0%
- 10
Qwen3.5-35B-A3B ↓5 0%
OS
12 models we track have a score here.
- 1 GPT-5.4 ↑1 73.3%
- 2 Kimi K2.6 ↓1 60%
- 3 Kimi K2.5 50%
- 4
Qwen3.5 397B-A17B 35%
- 5
Qwen3.5-35B-A3B 35%
- 6
Qwen 3.6 35B-A3B 20%
- 7
Qwen3-VL-235B-A22B ↑1 5.00%
- 8
UI-TARS-1.5 ↑4 0%
- 9 OpenCUA-32B 0%
- 10 OpenCUA-7B ↑1 0%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The DeskCraft paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official DeskCraft leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Computer and browser use
- BrowseComp In use Hard-to-find facts that need persistent web browsing. Short verifiable answers graded for exact correctness. Claude Fable 5 93.3%
- ScreenSpot-Pro In use GUI grounding on high-resolution professional-software screenshots. The model must locate the target UI element. GPT-6 Astra 92.7%
- AndroidWorld In use Agents complete 116 tasks across 20 real Android apps in an emulator. Each task checks the device's system state to judge success. GPT-6 Astra 100%
- OSWorld 2.1 In use Agents complete 108 long-horizon workflows across 31 self-hosted websites and desktop apps, averaging 318 tool calls. Scored by binary completion and checkpoint partial credit. Claude Opus 5 44.3%
- WebArena-Verified In use Agents complete 812 audited WebArena tasks on self-hosted websites. Deterministic evaluators check the agent's answer and captured network traces.
Muse Spark 1.1 69%
- OSWorld 2.1 (offline subset) In use Agents complete the 82 OSWorld 2.1 tasks on the official offline list, with no internet access. Scored by checkpoint partial credit and strict pass rate. Claude Fable 5.2 89.3%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 7 months ago, in April 2026, for Kimi K2.6. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the benchmark’s paper (18 entries).
- Pass Models scored 12 models we track, plus 6 other entries on its leaderboard.
- Warning Newest score The newest came 7 months ago, in April 2026, for Kimi K2.6.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.4 (#28).
- Pass Headroom The best is 29.8%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 12 rows for models we track and 6 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.