Grok 3: benchmark results
xAI's Grok 3 flagship model for reasoning and general tasks (February 2025). Provider: xAI. Released 2025-02-17. Access: API.
Aggregate Rating 1653 ± 6, rank #304 of 1656 rated models, from 982 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| IndicSafe | 83.67 | SAFE rate (%): share of the model's responses to the 6,000 I | 100 |
| IndicSafe - Unsafe Rate | 0.98 | UNSAFE rate (%): share of the model's responses to the 6,000 | 100 |
| LLM Emergent Collusion | 19 | Collusion Rate (%) | 100 |
| SRBench - Yelp (NDCG@5) | 0.05 | NDCG@5 (0-1) of the held-out next item among five recommenda | 100 |
| SRBench - Yelp (Recall@5) | 0.07 | Recall@5 (fraction) of the held-out next item among five rec | 100 |
| TaxPraBen - TaxCalc | 19 | Exact-match accuracy (%) on TaxCalc (tax payment calculation | 100 |
| Nejumi 4 - BFCL - Non-Live AST | 85.93 | Accuracy (%) | 99.3 |
| UGI Leaderboard | 63.47 | UGI Score | 99.3 |
| SpeechMap Compliance | 96.1 | % Requests Completed | 98 |
| ProLLM - Entity Extraction | 85.5 | Score (%) | 97.9 |
| EuroEval Swedish NLU - ScaLA SV | 71.91 | Matthews correlation coefficient (%) | 97.3 |
| EuroEval Icelandic NLU - Hotter and Colder Sentiment | 58.9 | Matthews correlation coefficient (%) | 97.2 |
Interactive version: theaggregate.ai/model?slug=grok-3 · How It Works · Data refreshed daily, snapshot 2026-10-20.