Is ScreenSpot-Pro still meaningful?
Status:Near ceiling Checked against the report, 5 November 2026
Verdict
Barely. Top models score 92.7%, and the test covers static screens only.
Why near ceiling: The chapter rates ScreenSpot-Pro Near ceiling. The best score rose from 18.9% at launch to 92.7% in November 2026.[1][1] ScreenSpot-Pro, arXiv 2504.07981 Primary[2][2] BenchLM, ScreenSpot-Pro Secondary
What “Near ceiling” meansKnown problems
- It covers static screens only, with no state tracking or recovery.[2][2] BenchLM, ScreenSpot-Pro Secondary
- All 2026 figures come from aggregators or labs. The Benchmarks chapter lists the same 92.7% as a vendor figure.[2][2] BenchLM, ScreenSpot-Pro Secondary[3][3] GPT-6 Astra benchmarks explained Vellum Secondary
What to look at instead
OSWorld-Verified tests real desktop tasks. Claude Fable 5 averaged 85% over five runs (aggregator).[4][4] BenchLM, Claude Fable 5 versus Opus 4.5 Secondary
- OSWorld-Verified Status:Near ceiling No longer for the frontier. Agents have reached the human baseline.
- OSWorld 2.0 Status:Still useful Yes. Full-credit scores stay low, though partial credit tells a kinder story.
Where it appears in the report
Chapter 1, Benchmarks
- Table 7. General and computer-use agent benchmarks, November 2026 in “Go deeper: General and computer-use agent benchmarks (18)”
Chapter 7, Multimodal, voice and embodied
- Figure 1. Tests Beyond Text Fill Up in Months, Too
- Table 1. Each modality has one number that tells its story
- GUI grounding has reached its ceiling
- Table 4. GUI grounding and computer-use scores in “Go deeper: Arena and GUI grounding detail”
- Figure 11. Robotics, GUI agents and deepfake detection have the weakest independent checks
Sources
Copied from the chapter source lists. Each number links back to the source in its chapter.
- [1]
- [2]
- [3] GPT-6 Astra benchmarks explained, Vellum. It transcribes OpenAI's launch table and discusses its framing. Secondary vellum.ai Chapter 1, source 14
- [4]