Grok 4.3: benchmark results
xAI's flagship Grok 4.3 multimodal model with native video input and a 1M-token context. Provider: xAI. Released 2026-04-30. Access: API.
Aggregate Rating 1704 ± 5, rank #188 of 1656 rated models, from 584 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA IFBench | 83.33 | Accuracy (%) | 100 |
| BLXBench | 85.5 | Score (imported) | 100 |
| CommunityFact (Closed-Input) - Finance (French) | 76.69 | Macro-F1 (%) over True and False claims on the temporally he | 100 |
| CommunityFact (Closed-Input) - Finance (Portuguese) | 73.3 | Macro-F1 (%) over True and False claims on the temporally he | 100 |
| CommunityFact (Closed-Input) - Finance (Spanish) | 72.37 | Macro-F1 (%) over True and False claims on the temporally he | 100 |
| CommunityFact (Closed-Input) - Politics (French) | 70.72 | Macro-F1 (%) over True and False claims on the temporally he | 100 |
| CommunityFact (Closed-Input) - Politics (Japanese) | 77.07 | Macro-F1 (%) over True and False claims on the temporally he | 100 |
| CommunityFact (Closed-Input) - Politics (Portuguese) | 73.24 | Macro-F1 (%) over True and False claims on the temporally he | 100 |
| PACT - Steerability | 0.56 | Score (0-1) | 100 |
| TAF-MED - Any-Turn Unsafe Rate | 35.6 | Any-turn unsafe rate (%; share of the 500 conversations with | 100 |
| TAF-MED - Collapse Rate | 24.4 | Collapse rate (%; share of the conversations with a strictly | 100 |
| TOC-Bench - Hallucination Diagnostic Accuracy | 53.8 | Hallucination diagnostic accuracy (%, HDA): accuracy on the | 100 |
Interactive version: theaggregate.ai/model?slug=grok-4-3 · How It Works · Data refreshed daily, snapshot 2026-10-20.