AI Model Capability Comparison
Computer use
Only models released within the past year are shown. Only benchmarks covering at least 3 of these models are shown. Unknown release dates are excluded; historical data is retained.
Agent's Last Exam — Pass rateAgent final exam — pass rate
— means no published score for this candidate in this source, not zero.3/11 recent-year models have scores
| No. | Model | |
|---|---|---|
| 1 |  Gemini 4 ArgonGemini 4 Argon | 39.5Pass rate |
| 2 |  Claude Opus 5.5Claude Opus 5.5 | 38.2Pass rate |
| 3 |  GPT-6 AstraGPT-6 Astra | 34.2Pass rate |
Google DeepMind (model card)
Source: Google DeepMind (model card) ↗
Google model card: publisher results and republished third-party scores, as of October 2026. See methodology for evaluation settings; these are not all independent evaluations. Evaluation methodology ↗
Agent's Last Exam — Pass rate · Source date unknown
PandaNpc Updated Oct 21, 2026, 06:01 UTC
Source: Epoch AI, CC BY 4.0 · Vals.ai · Artificial Analysis (artificialanalysis.ai)
Column guide
- No.
- Display order only. Sorting changes this number; it is not a published capability rank.
- Model
- The evaluated model name and its provider icon. A dash in a score column means no published score is available here, not a zero.
- Score
- Tests long-horizon professional work in real software environments. Pass rate counts tasks that meet the required outcome, not just partial progress. Evaluation methodology ↗
- Coverage
- Only published results for the selected model families are shown. OSWorld 2.x, Verified and Pro currently supply Claude Opus 5 configurations; this is not complete coverage of every model. Windows Agent Arena, MacArena, ScreenSpot-Pro, OSWorld-Human, WebArena-Verified, VisualWebArena, Online-Mind2Web and WorkArena++ are not yet integrated here. SafeArena measures safety separately and is not merged into task-completion scores. Missing results are never filled with zero.