← Back to directory
xAI
Grok 4
Released Jul 9, 2025 · knowledge cutoff 2024-11
- Status
- Superseded
- Location
- United States
- Type
- LLM
- Modality
- Multimodal
- Context window
- 256K
- Max output
- 64K
- Speed
- —
- Price ($/MTok in / out)
- $3 / $15
- Cost per task
- —
- Open weights
- No
Benchmarks7 of 10 reported
MMLU-Proretired
86.6%
GPQA Diamond
87.5%
SWE-bench Verified
72%
AIME (latest)retired
91.7%
Humanity's Last Exam
25.4%
LMArena Elo
1435
ARC-AGI-2
15.9%
SWE-bench Pro
—
Terminal-Bench 2.1
—
GDPval-AA v2
—
Strengths & weaknesses
Strengths
- Uses tools inside its reasoning rather than after it
- A heavy mode runs several attempts in parallel and picks the best for the hardest problems
- Genuinely strong on unfamiliar problems that can't be recalled
Weaknesses
- Very slow to first token because reasoning is always on
- Verbose reasoning that inflates cost on work that didn't need it
- Its behavior, not its capability, was the story at launch
For developers
API model strings
- xai
grok-4-0709
Licence
Proprietary — weights not released
Retirement
No retirement announced
Lineage
No recorded predecessor