Benchmark Almanac

DeskCraft

Kimi K2.6 29.8% leads since June 2026
Not in model cards

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.

Released
June 2026
Kind
Desktop apps
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Screens → Actions
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.

Abstract of DeskCraft: Benchmarking Desktop Agents on Professional Workflows and Human-in-the-Loop Collaboration (arXiv:2606.03103, CC0)
18
Entries in its paper, 12 models we track

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:

  1. 1 Kimi K2.6 29.8%
  2. 2 GPT-5.4 29.6%
  3. 3 Kimi K2.5 22.1%
  4. 4 Qwen3.5 397B-A17B 12.3%
  5. 5 Qwen3.5-35B-A3B 11.8%
  6. 6 Qwen 3.6 35B-A3B 8.55%
  7. 7 Qwen3-VL-32B 6.85%
  8. 8 Qwen3.5-9B 6.45%
  9. 9 EvoCUA-32B 5.70%
  10. 10 Qwen3-VL-235B-A22B 5.25%
  11. 11 OpenCUA-32B 4.80%
  12. 12 EvoCUA-8B 3.85%
  13. 13 OpenCUA-7B 2.95%
  14. 14 GUI-Owl-1.5-8B-Instruct 2.70%
  15. 15 Qwen3-VL-8B 2.65%
  16. 16 GUI-Owl-1.5-32B-Instruct 2.40%
  17. 17 UI-TARS-1.5 1.55%
  18. 18 OS-Atlas-Pro-7B 0.15%

Source: The DeskCraft paper

#EntryScoreSettingDateSource
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 Qwen3.5 397B-A17B 12.3% Feb 2026 Official leaderboard
5 Qwen3.5-35B-A3B 11.8% Feb 2026 Official leaderboard
6 Qwen 3.6 35B-A3B 8.55% Apr 2026 Official leaderboard
7 Qwen3-VL-32B 6.85% Official leaderboard
8 Qwen3.5-9B 6.45% Feb 2026 Official leaderboard
9 EvoCUA-32B 5.70% Official leaderboard
10 Qwen3-VL-235B-A22B 5.25% Sep 2025 Official leaderboard
11 OpenCUA-32B 4.80% Jul 2025 Official leaderboard
12 EvoCUA-8B 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 UI-TARS-1.5 1.55% Apr 2025 Official leaderboard
18 OS-Atlas-Pro-7B 0.15% Official leaderboard
#ModelScoreSettingDateSource
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 Qwen3.5 397B-A17B 12.3% Feb 2026 Official leaderboard
5 Qwen3.5-35B-A3B 11.8% Feb 2026 Official leaderboard
6 Qwen 3.6 35B-A3B 8.55% Apr 2026 Official leaderboard
7 Qwen3.5-9B 6.45% Feb 2026 Official leaderboard
8 Qwen3-VL-235B-A22B 5.25% Sep 2025 Official leaderboard
9 OpenCUA-32B 4.80% Jul 2025 Official leaderboard
10 EvoCUA-8B 3.85% Jan 2026 Official leaderboard
11 OpenCUA-7B 2.95% Jul 2025 Official leaderboard
12 UI-TARS-1.5 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. 1 Kimi K2.6 33.8%
  2. 2 GPT-5.4 31.6%
  3. 3 Kimi K2.5 20.3%
  4. 4 Qwen3.5 397B-A17B 12.9%
  5. 5 Qwen3.5-35B-A3B 11.1%
  6. 6 Qwen 3.6 35B-A3B 10.2%
  7. 7 OpenCUA-32B ↑2 9.60%
  8. 8 Qwen3.5-9B ↓1 8.70%
  9. 9 EvoCUA-8B ↑1 7.00%
  10. 10 OpenCUA-7B ↑1 5.20%

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Moonshot AI
Benchmark paper Not tested since Jun 2026
Kimi K2.6 29.8%

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.

Plot with other benchmarks
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Kimi K2.6 Kimi K2.5 Qwen3-VL-235B-A22B UI-TARS-1.5 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Kimi K2.6 Kimi K2.5

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

All 119 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.

How we check