Grok 4.5 (High): benchmark results
Grok 4.5 evaluated at the high reasoning-effort setting. Provider: xAI. Released 2026-07-08. Access: API.
Aggregate Rating 1807 ± 8, rank #126 of 2168 rated models, from 261 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Hy-MultiTurn - Constraint Synthesis | 90 | Importance-weighted score (%) on the 35 constraint synthesis | 100 |
| Hy-MultiTurn - Object Localization | 95.2 | Importance-weighted score (%) on the 35 object localization | 100 |
| Hy-MultiTurn - Reference Resolution | 78 | Importance-weighted score (%) on the 35 reference resolution | 100 |
| Nejumi 4 - HLE (Japanese) - Calibration Error | 34 | Calibration error (%) | 100 |
| Nejumi 4 - M-IFEval (Japanese) | 94.25 | Score (%) | 100 |
| xAI Grok 4.6 Model Card - Factuality (Hallucination) | 0.98 | Hallucination Rate (%, lower is better) | 100 |
| SWE-rebench | 63.78 | Resolved (%) | 99 |
| Nejumi 4 - BFCL - Relevance Detection | 100 | Accuracy (%) | 98.6 |
| ObviousBench | 100 | Answer pass³ (%) | 97.7 |
| Vals AI SkillsBench | 66.03 | Accuracy (%) | 97.1 |
| Nejumi 4 - jaster (2-shot) - JMMLU | 96 | Exact match (%) | 96.5 |
| Nejumi 4 - jaster (0-shot) - JCommonsenseQA | 100 | Exact match (%) | 96.2 |
Interactive version: theaggregate.ai/model?slug=grok-4-5-high · How It Works · Data refreshed daily, snapshot 2026-10-20.