Grok 4
X.AI·
LLMsflagshipcloud
Overview
Next generation model with enhanced understanding of the physical world.
Capabilities and innovations
Physical world modelVideo reasoningRobot control
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| GPQA Diamond | 88.1% | unsourced |
| Humanity’s Last Exam (no tools) | 33.2% | unsourced |
| SWE-bench Verified | 60.4% | unsourced |
| MMLU | 90.1% | unsourced |
| MMLU-Pro | 85.3% | unsourced |
| ARC-AGI-2 | 15.9% | independent Grok 4 (Thinking) set a new commercial SOTA of 15.9% on ARC-AGI-2 (Jul 2025), independently verified by ARC Prize on a blind holdout set. |
- SWE-bench Pro: not reported by the vendor
- Terminal-Bench 2.x: benchmark did not exist at release
- MMMU-Pro: benchmark did not exist at release
Links
More from X.AI
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.