Grok 4.20: benchmark results
xAI Grok 4.20 flagship with a built-in four-agent collaborative council and a 2M-token context. Provider: xAI. Released 2026-03-10. Access: API.
Aggregate Rating 1710 ± 5, rank #175 of 1656 rated models, from 585 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Alpha Arena (Nof1) | 34.59 | Return (%) | 100 |
| Causal Sensitivity Score (CSS) | 47.3 | CSS (self-reported) | 100 |
| ELSS - Landing Site Filtering (Nanjing) | 83.2 | F1 (%; the model judges each candidate landing site in a rem | 100 |
| EpiBench | 33.6 | Epitope Identification Accuracy (%, Task 2) | 100 |
| FutureSearch BTF-2 Calibration | 0 | Calibration Error | 100 |
| Quantum API Drift - Qiskit 1.3 | 51.33 | Pass@1 (%; unbiased estimator from 3 samples per task, macro | 100 |
| REFUTE | 74.5 | Truth Score (accuracy + calibration, higher is better) | 100 |
| VisualLeakBench - PII Propagation (Note Capture) | 72 | Tool propagation rate (%; non-error traces whose tool argume | 100 |
| BenchTable - Utility | 87.4 | Weighted Score (%) | 99.6 |
| SpeechMap Compliance | 98.2 | % Requests Completed | 99.1 |
| Horangi 4 - GLP - Mathematical Reasoning | 98 | Score (%) | 98.6 |
| Evals for Every Language - Japanese (ja) | 77.18 | Average Score (%) | 98.5 |
Interactive version: theaggregate.ai/model?slug=grok-4-20 · How It Works · Data refreshed daily, snapshot 2026-10-20.