Grok 4.1
X.AI·
LLMsflagshipcloud
Overview
Refined Grok 4 with improved safety and alignment.
Capabilities and innovations
Truth seekingReduced biasEnhanced safety
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| GPQA Diamond | 89.3% | unsourced |
| Humanity’s Last Exam (no tools) | 36.7% | unsourced |
| SWE-bench Verified | 62.1% | unsourced |
| MMLU | 91.2% | unsourced |
- SWE-bench Pro: not reported by the vendor
- MMLU-Pro: not reported by the vendor
- Terminal-Bench 2.x: benchmark did not exist at release
- MMMU-Pro: benchmark did not exist at release
Reliability
- Hallucination rate (Vectara HHEM)
- 19.2% (lower is better)independent
- Agentic tool use (τ-bench)
- 74.6% (higher is better)third party
Links
More from X.AI
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.