Benchmark Almanac

SCUBA

GPT-5 53.9% leads since September 2025
Not in model cards

Agents work in a Salesforce sandbox as administrators, sales representatives, or service agents. Evaluators check milestones such as creating accounts and routing support cases.

Salesforce compares nine agent configurations using seven distinct models on 300 tasks. TestZeus reports 107 passes among 122 tasks with written specifications.

Released
September 2025
Kind
Web browsing
Level
University
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Screens → Actions
Interaction
Agentic
Environment
Browser, Salesforce
Items from
Real work and use
Refresh
One fixed set
Run by
Anyone
Occupations
Computer and mathematical, Sales, Office and administrative support
Strongest model tested
GPT-5 #71 on the capability index
Last model added
August 2025
Scores refreshed
November 20th, 2026
8
Entries on its own leaderboard, 7 models we track

GPT-5 leads with 53.9% among models released by August 2025 GPT-5 leads with 53.9% among models released by August 2025

Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran on its own with several settings, we keep its best. Every run is in the next chart. Agents other companies built around a model get a pale shade of that model's lab color. Gray marks blends and entries not in our model list, such as fine-tunes and older models. None of these count toward leaders. Agents built around a model, such as Agent-S2.5 + GPT-5, are their own entries. We don't count them as the model's score. This view shows only models we track, each at its best.

The newest model with a score here came out in August 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-5 demonstration 53.9%
  2. 2 o3 demonstration 50%
  3. 3 Claude Sonnet 4 demonstration 47.7%
  4. 4 Gemini 2.5 Pro Jun 2025, demonstration 46.1%
  5. 5 Agent-S2.5 + GPT-5 demonstration 40.4%
  6. 6 OpenAI Computer-Using Agent (CUA) demonstration 28.9%
  7. 7 UI-TARS-1.5 demonstration 9.16%
  8. 8 OpenCUA-7B demonstration 5.38%

Source: Official SCUBA leaderboard

#EntryScoreSettingDateSource
1 GPT-5 53.9% demonstration, browser Aug 2025 Official leaderboard
2 o3 50% demonstration, browser Apr 2025 Official leaderboard
3 Claude Sonnet 4 47.7% demonstration, computer May 2025 Official leaderboard
4 Gemini 2.5 Pro Jun 2025 46.1% demonstration, browser Jun 2025 Official leaderboard
5 Agent-S2.5 + GPT-5 40.4% demonstration Aug 2025 Official leaderboard
6 OpenAI Computer-Using Agent (CUA) 28.9% demonstration, computer Jan 2025 Official leaderboard
7 UI-TARS-1.5 9.16% demonstration, computer Apr 2025 Official leaderboard
8 OpenCUA-7B 5.38% demonstration, computer Jul 2025 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-5 53.9% demonstration, browser Aug 2025 Official leaderboard
2 o3 50% demonstration, browser Apr 2025 Official leaderboard
3 Claude Sonnet 4 47.7% demonstration, computer May 2025 Official leaderboard
4 Gemini 2.5 Pro Jun 2025 46.1% demonstration, browser Jun 2025 Official leaderboard
5 OpenAI Computer-Using Agent (CUA) 28.9% demonstration, computer Jan 2025 Official leaderboard
6 UI-TARS-1.5 9.16% demonstration, computer Apr 2025 Official leaderboard
7 OpenCUA-7B 5.38% demonstration, computer Jul 2025 Official leaderboard

The best score, 53.9%, costs $0.51 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Google DeepMind OpenAI Cost frontier One model at each effort level
0% 20% 40% 60% $0.10 $0.20 $0.50 $1.00 $2.00 $5.00 $10 Cost per task, log scale GPT-5, demonstration Gemini 2.5 Pro, demonstration Gemini 2.5 Pro, zero-shot 0% 20% 40% 60% $0.10 $1.00 $10 Cost per task, log scale GPT-5, demonstration Gemini 2.5 Pro, demonstration

Source: Official SCUBA leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5 demonstration browser 53.9% $0.51
GPT-5 zero-shot browser 51.3% $0.55
o3 demonstration browser 50% $0.55
Claude Sonnet 4 demonstration computer 47.7% $1.34
Claude Sonnet 4 demonstration browser 46.1% $1.19
Gemini 2.5 Pro Jun 2025 demonstration browser 46.1% $0.31
o3 zero-shot browser 45.7% $0.56
Agent-S2.5 + GPT-5 demonstration Agent-S2.5 40.4% $1.19
Agent-S2.5 + GPT-5 zero-shot Agent-S2.5 39% $1.17
Claude Sonnet 4 zero-shot browser 34.7% $1.16
Gemini 2.5 Pro Jun 2025 zero-shot browser 31% $0.24
OpenAI Computer-Using Agent (CUA) demonstration computer 28.9% $0.52
Claude Sonnet 4 zero-shot computer 27% $1.44
OpenAI Computer-Using Agent (CUA) zero-shot computer 16% $0.59
UI-TARS-1.5 demonstration computer 9.16%
OpenCUA-7B demonstration computer 5.38%
UI-TARS-1.5 zero-shot computer 2.67%
OpenCUA-7B zero-shot computer 0.67%

The best score, 53.9%, costs $0.51 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Google DeepMind OpenAI Cost frontier One model at each effort level
0% 20% 40% 60% $0.10 $0.20 $0.50 $1.00 $2.00 $5.00 $10 Cost per task, log scale GPT-5, demonstration Gemini 2.5 Pro, demonstration Gemini 2.5 Pro, zero-shot 0% 20% 40% 60% $0.10 $1.00 $10 Cost per task, log scale GPT-5, demonstration Gemini 2.5 Pro, demonstration

Source: Official SCUBA leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5 demonstration browser 53.9% $0.51
GPT-5 zero-shot browser 51.3% $0.55
o3 demonstration browser 50% $0.55
Claude Sonnet 4 demonstration computer 47.7% $1.34
Claude Sonnet 4 demonstration browser 46.1% $1.19
Gemini 2.5 Pro Jun 2025 demonstration browser 46.1% $0.31
o3 zero-shot browser 45.7% $0.56
Claude Sonnet 4 zero-shot browser 34.7% $1.16
Gemini 2.5 Pro Jun 2025 zero-shot browser 31% $0.24
OpenAI Computer-Using Agent (CUA) demonstration computer 28.9% $0.52
Claude Sonnet 4 zero-shot computer 27% $1.44
OpenAI Computer-Using Agent (CUA) zero-shot computer 16% $0.59
UI-TARS-1.5 demonstration computer 9.16%
OpenCUA-7B demonstration computer 5.38%
UI-TARS-1.5 zero-shot computer 2.67%
OpenCUA-7B zero-shot computer 0.67%

Number one over time

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

OpenAI
Own leaderboard Not tested since Sep 2025
GPT-5 53.9%

Source: Official SCUBA leaderboard

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
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) GPT-5 o3 OpenAI Computer-Using Agent (CUA) 0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) GPT-5 OpenAI Computer-Using Agent (CUA)

Source: Official SCUBA 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 15 months ago, in August 2025, for GPT-5. 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 official leaderboard (8 entries).
  • Pass Models scored 7 models we track, plus 1 other entry on its leaderboard.
  • Warning Newest score The newest came 15 months ago, in August 2025, for GPT-5.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5 (#71).
  • Pass Headroom The best is 53.9%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on sfrcua.github.io: 18 rows for models we track 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