Benchmark Almanac

ScreenSpot-Pro

GPT-6 Astra 92.7% leads since September 2026
UI-TARS-1.5 61.6% leads since April 2025
GPT-6 Astra 92.7% leads since September 2026
In use

GUI grounding on high-resolution professional-software screenshots. The model must locate the target UI element.

Astra score known only from secondary write-ups of OpenAI launch materials.

Released
January 2025
Kind
Finding things on screen
Level
University
Scored
Checked automatically
Answers
Short answer
Tasks
All public
Strongest model tested
GPT-6 Astra #3 on the capability index
Last in a launch post
November 2026
Last model added
November 2026
Scores refreshed
November 12th, 2026

In the authors’ words

Recent advancements in Multi-modal Large Language Models (MLLMs) have led to significant progress in developing GUI agents for general tasks such as web browsing and mobile phone use. However, their application in professional domains remains under-explored. These specialized workflows introduce unique challenges for GUI perception models, including high-resolution displays, smaller target sizes, and complex environments. In this paper, we introduce ScreenSpot-Pro, a new benchmark designed to rigorously evaluate the grounding capabilities of MLLMs in high-resolution professional settings. The benchmark comprises authentic high-resolution images from a variety of professional domains with expert annotations. It spans 23 applications across five industries and three operating systems. Existing GUI grounding models perform poorly on this dataset, with the best model achieving only 18.9%. Our experiments reveal that strategically reducing the search area enhances accuracy. Based on this insight, we propose ScreenSeekeR, a visual search method that utilizes the GUI knowledge of a strong planner to guide a cascaded search, achieving state-of-the-art performance with 48.1% without any additional training. We hope that our benchmark and findings will advance the development of GUI agents for professional applications. Code, data and leaderboard can be found at https://gui-agent.github.io/grounding-leaderboard.

Abstract of ScreenSpot-Pro: GUI Grounding for Professional High-Resolution Computer Use (arXiv:2504.07981, CC0)
32
Models scored in lab cards
27
Launch posts list it, from 9 labs
14
System cards report it, from 8 labs

GPT-6 Astra leads with 92.7%

Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.

UI-TARS-1.5 leads with 61.6% across all independent runs

Best score per model across the official leaderboard, Epoch AI’s own runs and any board Epoch AI copied. Each bar and row names its source.

GPT-6 Astra leads with 92.7%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.

32 models. Scroll sideways for the rest.

  1. 92.7 GPT-6 Astra
  2. 87.9 Claude Opus 4.8
  3. 87.6 Claude Opus 4.7
  4. 86.3 GPT-5.2
  5. 84.5 Qwen 3.8 Max
  6. 84.1 Muse Spark
  7. 83.8 Inkling 2 Small
  8. 79.5 Claude Mythos Preview
  9. 79.0 Qwen3.7-Plus
  10. 75.4 Muse Glimmer
  11. 73.1 Seed1.8
  12. 72.7 Gemini 3 Pro
  13. 70.4 Qwen3.5-122B-A10B
  14. 70.3 Qwen3.5-27B
  15. 69.1 Gemini 3 Flash
  16. 68.6 Qwen3.5-35B-A3B
  17. 68.2 Qwen 3.6 Plus
  18. 66.8 Fara1.5-27B
  19. 65.6 Qwen3.5 397B-A17B
  20. 65.2 Qwen3.5-9B
  21. 61.8 Qwen3-VL-235B-A22B
  22. 61.6 UI-TARS-1.5
  23. 60.9 Seed1.5-VL
  24. 60.3 Qwen3.5-4B
  25. 57.8 Nemotron 3 Nano Omni 30B-A3B
  26. 54.5 Qwen3.5-2B
  27. 52.8 Kimi-VL-A3B-Thinking-2506
  28. 51.5 STEP3-VL-10B
  29. 46.5 Qwen3.5-0.8B
  30. 43.6 Qwen2.5-VL-72B
  31. 43.5 Nemotron 3.5 Super VL
  32. 5.50 Nemotron Nano 12B v2 VL
  1. 1 GPT-6 Astra 92.7%
  2. 2 Claude Opus 4.8 87.9%
  3. 3 Claude Opus 4.7 87.6%
  4. 4 GPT-5.2 86.3%
  5. 5 Qwen 3.8 Max 84.5%
  6. 6 Muse Spark 84.1%
  7. 7 Inkling 2 Small 83.8%
  8. 8 Claude Mythos Preview 79.5%
  9. 9 Qwen3.7-Plus 79%
  10. 10 Muse Glimmer 75.4%
  11. 11 Seed1.8 73.1%
  12. 12 Gemini 3 Pro 72.7%
  13. 13 Qwen3.5-122B-A10B 70.4%
  14. 14 Qwen3.5-27B 70.3%
  15. 15 Gemini 3 Flash 69.1%
  16. 16 Qwen3.5-35B-A3B 68.6%
  17. 17 Qwen 3.6 Plus 68.2%
  18. 18 Fara1.5-27B 66.8%
  19. 19 Qwen3.5 397B-A17B 65.6%
  20. 20 Qwen3.5-9B 65.2%
  21. 21 Qwen3-VL-235B-A22B 61.8%
  22. 22 UI-TARS-1.5 61.6%
  23. 23 Seed1.5-VL 60.9%
  24. 24 Qwen3.5-4B 60.3%
  25. 25 Nemotron 3 Nano Omni 30B-A3B 57.8%
  26. 26 Qwen3.5-2B 54.5%
  27. 27 Kimi-VL-A3B-Thinking-2506 52.8%
  28. 28 STEP3-VL-10B 51.5%
  29. 29 Qwen3.5-0.8B 46.5%
  30. 30 Qwen2.5-VL-72B 43.6%
  31. 31 Nemotron 3.5 Super VL 43.5%
  32. 32 Nemotron Nano 12B v2 VL 5.50%

Source: Labs’ model cards

Source: Official ScreenSpot-Pro leaderboard

39 models. Scroll sideways for the rest.

  1. 92.7 GPT-6 Astra
  2. 87.9 Claude Opus 4.8
  3. 87.6 Claude Opus 4.7
  4. 86.3 GPT-5.2
  5. 84.5 Qwen 3.8 Max
  6. 84.1 Muse Spark
  7. 83.8 Inkling 2 Small
  8. 79.5 Claude Mythos Preview
  9. 79.0 Qwen3.7-Plus
  10. 75.4 Muse Glimmer
  11. 73.1 Seed1.8
  12. 72.7 Gemini 3 Pro
  13. 70.4 Qwen3.5-122B-A10B
  14. 70.3 Qwen3.5-27B
  15. 69.1 Gemini 3 Flash
  16. 68.6 Qwen3.5-35B-A3B
  17. 68.2 Qwen 3.6 Plus
  18. 66.8 Fara1.5-27B
  19. 65.6 Qwen3.5 397B-A17B
  20. 65.2 Qwen3.5-9B
  21. 61.8 Qwen3-VL-235B-A22B
  22. 61.6 UI-TARS-1.5
  23. 60.9 Seed1.5-VL
  24. 60.3 Qwen3.5-4B
  25. 57.8 Nemotron 3 Nano Omni 30B-A3B
  26. 54.5 Qwen3.5-2B
  27. 53.3 Qwen2.5-VL-72B
  28. 52.8 Kimi-VL-A3B-Thinking-2506
  29. 51.5 STEP3-VL-10B
  30. 48.0 Qwen2.5-VL-32B
  31. 46.5 Qwen3.5-0.8B
  32. 43.5 Nemotron 3.5 Super VL
  33. 26.8 Qwen2.5-VL-7B
  34. 18.5 GPT-5
  35. 16.1 Qwen2.5-VL-3B
  36. 7.72 CogAgent
  37. 5.50 Nemotron Nano 12B v2 VL
  38. 1.58 Qwen2-VL-7B-Instruct
  39. 1.01 Qwen2-VL-72B-Instruct
  1. 1 GPT-6 Astra 92.7%
  2. 2 Claude Opus 4.8 87.9%
  3. 3 Claude Opus 4.7 87.6%
  4. 4 GPT-5.2 86.3%
  5. 5 Qwen 3.8 Max 84.5%
  6. 6 Muse Spark 84.1%
  7. 7 Inkling 2 Small 83.8%
  8. 8 Claude Mythos Preview 79.5%
  9. 9 Qwen3.7-Plus 79%
  10. 10 Muse Glimmer 75.4%
  11. 11 Seed1.8 73.1%
  12. 12 Gemini 3 Pro 72.7%
  13. 13 Qwen3.5-122B-A10B 70.4%
  14. 14 Qwen3.5-27B 70.3%
  15. 15 Gemini 3 Flash 69.1%
  16. 16 Qwen3.5-35B-A3B 68.6%
  17. 17 Qwen 3.6 Plus 68.2%
  18. 18 Fara1.5-27B 66.8%
  19. 19 Qwen3.5 397B-A17B 65.6%
  20. 20 Qwen3.5-9B 65.2%
  21. 21 Qwen3-VL-235B-A22B 61.8%
  22. 22 UI-TARS-1.5 61.6%
  23. 23 Seed1.5-VL 60.9%
  24. 24 Qwen3.5-4B 60.3%
  25. 25 Nemotron 3 Nano Omni 30B-A3B 57.8%
  26. 26 Qwen3.5-2B 54.5%
  27. 27 Qwen2.5-VL-72B 53.3%
  28. 28 Kimi-VL-A3B-Thinking-2506 52.8%
  29. 29 STEP3-VL-10B 51.5%
  30. 30 Qwen2.5-VL-32B 48%
  31. 31 Qwen3.5-0.8B 46.5%
  32. 32 Nemotron 3.5 Super VL 43.5%
  33. 33 Qwen2.5-VL-7B 26.8%
  34. 34 GPT-5 minimal 18.5%
  35. 35 Qwen2.5-VL-3B 16.1%
  36. 36 CogAgent 7.72%
  37. 37 Nemotron Nano 12B v2 VL 5.50%
  38. 38 Qwen2-VL-7B-Instruct 1.58%
  39. 39 Qwen2-VL-72B-Instruct 1.01%

Sources: Labs’ model cards and official ScreenSpot-Pro leaderboard

#ModelScoreSettingDateSource
1 GPT-6 Astra 92.7% Sep 2026 Lab model card
2 Claude Opus 4.8 87.9% May 2026 Lab model card
3 Claude Opus 4.7 87.6% Apr 2026 Lab model card
4 GPT-5.2 86.3% Dec 2025 Lab model card
5 Qwen 3.8 Max 84.5% Aug 2026 Lab model card
6 Muse Spark 84.1% Apr 2026 Lab model card
7 Inkling 2 Small 83.8% Nov 2026 Lab model card
8 Claude Mythos Preview 79.5% Apr 2026 Lab model card
9 Qwen3.7-Plus 79% Jun 2026 Lab model card
10 Muse Glimmer 75.4% Aug 2026 Lab model card
11 Seed1.8 73.1% Dec 2025 Lab model card
12 Gemini 3 Pro 72.7% Nov 2025 Lab model card
13 Qwen3.5-122B-A10B 70.4% Feb 2026 Lab model card
14 Qwen3.5-27B 70.3% Feb 2026 Lab model card
15 Gemini 3 Flash 69.1% Dec 2025 Lab model card
16 Qwen3.5-35B-A3B 68.6% Feb 2026 Lab model card
17 Qwen 3.6 Plus 68.2% Apr 2026 Lab model card
18 Fara1.5-27B 66.8% Jun 2026 Lab model card
19 Qwen3.5 397B-A17B 65.6% Feb 2026 Lab model card
20 Qwen3.5-9B 65.2% Mar 2026 Lab model card
21 Qwen3-VL-235B-A22B 61.8% Sep 2025 Lab model card
22 UI-TARS-1.5 61.6% Apr 2025 Lab model card
23 Seed1.5-VL 60.9% May 2025 Lab model card
24 Qwen3.5-4B 60.3% Mar 2026 Lab model card
25 Nemotron 3 Nano Omni 30B-A3B 57.8% Apr 2026 Lab model card
26 Qwen3.5-2B 54.5% Mar 2026 Lab model card
27 Kimi-VL-A3B-Thinking-2506 52.8% Jun 2025 Lab model card
28 STEP3-VL-10B 51.5% Jan 2026 Lab model card
29 Qwen3.5-0.8B 46.5% Mar 2026 Lab model card
30 Qwen2.5-VL-72B 43.6% Jan 2025 Lab model card
31 Nemotron 3.5 Super VL 43.5% Oct 2026 Lab model card
32 Nemotron Nano 12B v2 VL 5.50% Nov 2025 Lab model card
#ModelScoreSettingDateSource
1 UI-TARS-1.5 61.6% Apr 2025 Official leaderboard
2 Seed1.5-VL 60.9% May 2025 Official leaderboard
3 Qwen2.5-VL-72B 53.3% Jan 2025 Official leaderboard
4 Kimi-VL-A3B-Thinking-2506 51% Jun 2025 Official leaderboard
5 Qwen2.5-VL-32B 48% Mar 2025 Official leaderboard
6 Qwen2.5-VL-7B 26.8% Jan 2025 Official leaderboard
7 GPT-5 18.5% minimal, resized screenshots Aug 2025 Official leaderboard
8 Qwen2.5-VL-3B 16.1% Jan 2025 Official leaderboard
9 CogAgent 7.72% Dec 2023 Official leaderboard
10 Qwen2-VL-7B-Instruct 1.58% Aug 2024 Official leaderboard
11 Qwen2-VL-72B-Instruct 1.01% Aug 2024 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-6 Astra 92.7% Sep 2026 Lab model card
2 Claude Opus 4.8 87.9% May 2026 Lab model card
3 Claude Opus 4.7 87.6% Apr 2026 Lab model card
4 GPT-5.2 86.3% Dec 2025 Lab model card
5 Qwen 3.8 Max 84.5% Aug 2026 Lab model card
6 Muse Spark 84.1% Apr 2026 Lab model card
7 Inkling 2 Small 83.8% Nov 2026 Lab model card
8 Claude Mythos Preview 79.5% Apr 2026 Lab model card
9 Qwen3.7-Plus 79% Jun 2026 Lab model card
10 Muse Glimmer 75.4% Aug 2026 Lab model card
11 Seed1.8 73.1% Dec 2025 Lab model card
12 Gemini 3 Pro 72.7% Nov 2025 Lab model card
13 Qwen3.5-122B-A10B 70.4% Feb 2026 Lab model card
14 Qwen3.5-27B 70.3% Feb 2026 Lab model card
15 Gemini 3 Flash 69.1% Dec 2025 Lab model card
16 Qwen3.5-35B-A3B 68.6% Feb 2026 Lab model card
17 Qwen 3.6 Plus 68.2% Apr 2026 Lab model card
18 Fara1.5-27B 66.8% Jun 2026 Lab model card
19 Qwen3.5 397B-A17B 65.6% Feb 2026 Lab model card
20 Qwen3.5-9B 65.2% Mar 2026 Lab model card
21 Qwen3-VL-235B-A22B 61.8% Sep 2025 Lab model card
22 UI-TARS-1.5 61.6% Apr 2025 Official leaderboard
23 Seed1.5-VL 60.9% May 2025 Official leaderboard
24 Qwen3.5-4B 60.3% Mar 2026 Lab model card
25 Nemotron 3 Nano Omni 30B-A3B 57.8% Apr 2026 Lab model card
26 Qwen3.5-2B 54.5% Mar 2026 Lab model card
27 Qwen2.5-VL-72B 53.3% Jan 2025 Official leaderboard
28 Kimi-VL-A3B-Thinking-2506 52.8% Jun 2025 Lab model card
29 STEP3-VL-10B 51.5% Jan 2026 Lab model card
30 Qwen2.5-VL-32B 48% Mar 2025 Official leaderboard
31 Qwen3.5-0.8B 46.5% Mar 2026 Lab model card
32 Nemotron 3.5 Super VL 43.5% Oct 2026 Lab model card
33 Qwen2.5-VL-7B 26.8% Jan 2025 Official leaderboard
34 GPT-5 18.5% minimal, resized screenshots Aug 2025 Official leaderboard
35 Qwen2.5-VL-3B 16.1% Jan 2025 Official leaderboard
36 CogAgent 7.72% Dec 2023 Official leaderboard
37 Nemotron Nano 12B v2 VL 5.50% Nov 2025 Lab model card
38 Qwen2-VL-7B-Instruct 1.58% Aug 2024 Official leaderboard
39 Qwen2-VL-72B-Instruct 1.01% Aug 2024 Official leaderboard

The benchmark's own leaderboard has 98 entries, 11 of them models we track. Switch the scores above to rank them together.

8 models have held number one in labs’ own numbers since its release in January 2025 3 models have held number one across all independent runs since its release in January 2025 9 models have held number one with claims included since its release in January 2025

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

Z.ai Qwen ByteDance Seed Google DeepMind OpenAI Anthropic
Own leaderboard Not tested since Aug 2025
UI-TARS-1.5 61.6%
Lab claims Passed 85% in Dec 2025, by GPT-5.2
GPT-6 Astra 92.7%

Sources: Official ScreenSpot-Pro leaderboard and labs’ model cards

OpenAI’s own number for GPT-5.2 passed 85% in December 2025

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2024 2025 2026 2027 Near the ceiling (85%) GPT-6 Astra Claude Opus 4.8 Gemini 3 Pro Qwen3-VL-235B-A22B Qwen2.5-VL-72B CogAgent 0% 20% 40% 60% 80% 100% 2024 2025 2026 2027 Near the ceiling (85%) GPT-6 Astra GPT-5.2

Sources: Official ScreenSpot-Pro leaderboard and labs’ model cards

4 labs put it in a launch post in the last six months

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

Qwen
ByteDance Seed
Moonshot AI
Google DeepMind
OpenAI
StepFun
Meta
Anthropic
Thinking Machines
DateLaunch postScoreAs the post writes it
Nov 5, 2026 Inkling 2 Small 83.8 ScreenSpot Pro (original; 91.5 with Python)
Sep 3, 2026 GPT-6 Astra 92.7 ScreenSpot-Pro (no tools)
Aug 10, 2026 Muse Glimmer 75.4 ScreenSpot Pro
Aug 3, 2026 Qwen 3.8 Max 84.5 ScreenSpot Pro
Jun 1, 2026 Qwen3.7-Plus 79 ScreenSpot Pro (enable_thinking=False)
May 28, 2026 Claude Opus 4.8 87.9 ScreenSpot-Pro (with tools)
Apr 16, 2026 Claude Opus 4.7 87.6 ScreenSpot-Pro (with tools)
Apr 8, 2026 Muse Spark 84.1 ScreenSpot Pro (Screenshot Localization - With Python)
Apr 2, 2026 Qwen 3.6 Plus 68.2 ScreenSpot Pro
Mar 2, 2026 Qwen3.5-0.8B 46.5 ScreenSpot Pro --/46.5
Mar 2, 2026 Qwen3.5-2B 54.5 ScreenSpot Pro --/54.5
Mar 2, 2026 Qwen3.5-4B 60.3 ScreenSpot Pro
Mar 2, 2026 Qwen3.5-9B 65.2 ScreenSpot Pro
Feb 24, 2026 Qwen3.5-35B-A3B 68.6 ScreenSpot Pro
Feb 24, 2026 Qwen3.5-27B 70.3 ScreenSpot Pro
Feb 24, 2026 Qwen3.5-122B-A10B 70.4 ScreenSpot Pro
Feb 16, 2026 Qwen3.5 397B-A17B 65.6 ScreenSpot Pro
Jan 13, 2026 STEP3-VL-10B 51.5 ScreenSpot-Pro
Dec 18, 2025 Seed1.8 73.1 ScreenSpot-Pro 64.3 / 73.1 with crop-box tool
Dec 17, 2025 Gemini 3 Flash 69.1 ScreenSpot-Pro (No tools unless specified)
Dec 11, 2025 GPT-5.2 86.3 ScreenSpot-Pro GUI screenshot understanding (w/ Python)
Nov 18, 2025 Gemini 3 Pro 72.7 ScreenSpot-Pro
Sep 23, 2025 Qwen3-VL-235B-A22B 61.8 ScreenSpot Pro
Jun 21, 2025 Kimi-VL-A3B-Thinking-2506 52.8 ScreenSpot-Pro (Acc)
May 12, 2025 Seed1.5-VL 60.9 ScreenSpot-Pro
Apr 16, 2025 UI-TARS-1.5 61.6 ScreenSpotPro
Jan 28, 2025 Qwen2.5-VL-72B 43.6 ScreenSpot Pro

14 system cards report it, from 8 labs

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Oct 19, 2026 Muse Glimmer technical report 75.4 Screenspot Pro, high (Tables 2 and 5) Capabilities
Oct 15, 2026 Nemotron 3.5 Super VL technical report 43.5 ScreenSpot Pro, BF16 acc. (Table 23) Capabilities
Jun 18, 2026 Fara1.5-27B technical report 66.8 ScreenSpot-Pro with two-step zoom, 58.6 without zoom (Table 6) Capabilities
May 28, 2026 Claude Opus 4.8 system card 82.3 ScreenSpot-Pro, no tools, max effort (87.9% with Python tools) (Table 8.1.A, 8.12.5) Capabilities
Apr 27, 2026 Nemotron 3 Nano Omni 30B-A3B technical report 57.8 ScreenSpot-Pro, reasoning on, 59.3 reasoning off (Table 7) Capabilities
Apr 16, 2026 Claude Opus 4.7 system card 79.5 ScreenSpot-Pro, no tools. 87.6 with Python tools (Table 8.1.A, 8.9.3) Capabilities
Apr 7, 2026 Claude Mythos Preview system card 79.5 ScreenSpot-Pro, adaptive thinking, max effort, no tools, average of five runs (6.11.2); 92.8 with Python tools Capabilities
Mar 21, 2026 Seed1.8 model card 64.3 ScreenSpot-Pro 64.3, 73.1 with the crop-box tool (Table 2) Capabilities
Jan 14, 2026 STEP3-VL-10B technical report 51.5 ScreenSpot-Pro, SeRe (Table 1) Capabilities
Nov 26, 2025 Qwen3-VL-235B-A22B technical report 61.8 ScreenSpot Pro (Table 2) Capabilities
Nov 6, 2025 Nemotron Nano 12B v2 VL technical report 5.5 ScreenSpot-Pro, reasoning on (Table 3). Reasoning off 4.8 Capabilities
Jun 23, 2025 Kimi-VL-A3B-Thinking-2506 technical report 52.8 ScreenSpot-Pro (Acc) (Table 5) Capabilities
May 11, 2025 Seed1.5-VL technical report 60.9 ScreenSpot-Pro, thinking (Table 8) Capabilities
Feb 19, 2025 Qwen2.5-VL-72B technical report 43.6 ScreenSpot Pro (Table 9) Capabilities

The lab's number comes out ahead in 1 of 2 pairs that differ

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. 2 of them match, which usually means one side copied the other. Every benchmark.

Model Lab says Board says Gap
Qwen2.5-VL-72B 43.6% ScreenSpot Pro 53.3% Leaderboard −9.7 pts
Kimi-VL-A3B-Thinking-2506 52.8% ScreenSpot-Pro (Acc) 51% Leaderboard +1.8 pts
UI-TARS-1.5 61.6% ScreenSpotPro 61.6% Leaderboard Same
Seed1.5-VL 60.9% ScreenSpot-Pro 60.9% Leaderboard Same

More in Computer and browser use

All 69 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%
  • 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 5.5 81.8%
  • 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%
  • MobileWorld In use Agents complete 201 tasks across 20 mobile apps, including tasks that need user clarification or MCP tools. Backend checks verify success. GPT-5 55.3%
  • RecreationBench In use Agents explore a running reference app and rebuild it, across 250 tasks on five platforms. Programmatic tests and a VLM judge score the result. GPT-6 Astra 58.1%