Grok 2 Vision: benchmark results
Provider: xAI. Access: API.
Aggregate Rating 1566 ± 26, rank #1462 of 10916 rated models, from 44 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MA-Bench (Open-Ended) | 0.65 | GPT-4o judge score (0 to 5) averaged over the six open-ended | 85.7 |
| VLRS-Bench - Mechanistic Interaction Reasoning | 36.8 | Score (%) on the 125 Mechanistic Interaction Reasoning (sing | 81.8 |
| MA-Bench - Fine-Grained Recognition | 31.3 | Accuracy (%) on the fine-grained (action-level) micro-action | 81 |
| MA-Bench - Coarse-Grained Recognition | 19.3 | Accuracy (%) on the coarse-grained (body-level) micro-action | 61.9 |
| ContrAR (Multiple Frames) - Safety Inspection | 76.19 | Detection accuracy (%) for contradictory virtual content att | 60 |
| MA-Bench (Closed-Ended) | 43.25 | Accuracy (%) averaged over the six closed-ended tasks (coars | 57.1 |
| ContrAR (Multiple Frames) - Indoor Navigation | 74.07 | Detection accuracy (%) for contradictory virtual content att | 50 |
| ProLLM - Image Understanding | 60.4 | Score (%) | 50 |
| AI for Education Visual Reasoning - match (process) | 36.1 | Accuracy (%) | 49.2 |
| VLRS-Bench - Spatiotemporal Consistency Reasoning | 28 | Score (%) on the 125 Spatiotemporal Consistency Reasoning (t | 42.1 |
| ContrAR (Multiple Frames) | 66.03 | Detection accuracy (%) for contradictory virtual content att | 40 |
| ContrAR (Multiple Frames) - Outdoor Navigation | 73.33 | Detection accuracy (%) for contradictory virtual content att | 40 |
Interactive version: theaggregate.ai/model?slug=grok-2-vision · How It Works · Data refreshed daily, snapshot 2026-11-08.