Grok 3 (xAI) and Llama 4 Behemoth (preview) (Meta AI) are both frontier AI models. Grok 3 was published in February 2025 and Llama 4 Behemoth (preview) in April 2025.
Grok 3 was trained on 3.5 x 10^26 FLOP, about 6.8x the compute of Llama 4 Behemoth (preview) at 5.2 x 10^25 FLOP. Training compute is the closest available proxy for how much was invested in a model, though it says nothing on its own about how well that compute was spent.
These two models share no benchmark on which both have been scored, so no direct performance comparison is possible here. The specification table below is a comparison of inputs, not of results.
| Field | Grok 3 | Llama 4 Behemoth (preview) |
|---|---|---|
| Organization | xAI | Meta AI |
| Published | Feb 17, 2025 | Apr 5, 2025 |
| Training compute | 3.5 x 10^26 FLOP | 5.2 x 10^25 FLOP |
| Parameters | 3T | 2T |
| Dataset size | — | 30T |
| Training hardware | NVIDIA H100 SXM5 80GB | NVIDIA H100 SXM5 80GB |
| Chips used | 80,000 | 32,000 |
| Training time | 2.2K h | — |
| Training cost (2023 USD) | $218M | $45M |
| Training power draw | 109.9 MW | 43.9 MW |
| Accessibility | API access | Unreleased |
| Open weights | No | No |
| Country | United States | United States |
