Grok-2 and Grok-2 mini
Grok-2 and Grok-2 mini launch in beta on X with large gains over Grok-1.5 and image generation via FLUX.1.
Frontier-class chat, coding and reasoning for xAI: MMLU 87.5%, GPQA 56.0%, MATH 76.1%, HumanEval 88.4%. Image generation was handled by Black Forest Labs' FLUX.1, not an in-house model. Enterprise API followed in Dec 2024 (grok-2-1212).
- Date
- Tuesday, 13 August 2024
- Lab
- xAI
- Kind
- model
- Access
- app only
Figures
| Measure | Value | Measured by |
|---|---|---|
| MMLU | 87.5% Grok-2 mini 86.2% | company |
| GPQA | 56.0% Grok-2 mini 51.0% | company |
| MATH | 76.1% Grok-2 mini 73.0% | company |
| HumanEval | 88.4% Grok-2 mini 85.7% | company |
Wikipedia lists a different date (2024-08-20); xAI's post is dated 2024-08-13. Weights were open-sourced a year later as 'Grok 2.5' (see xai-grok-2-5-open-weights).
Sources
This record was checked against its sources on 6 October 2026. How we check