GPT-4 (Mar 2023) (OpenAI) and Llama 3.1-405B (Meta AI) are both frontier AI models. GPT-4 (Mar 2023) was published in March 2023 and Llama 3.1-405B in July 2024.
Llama 3.1-405B was trained on 3.8 x 10^25 FLOP, about 1.8x the compute of GPT-4 (Mar 2023) at 2.1 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.
Llama 3.1-405B wins the single benchmark both were scored on. Benchmark counts are a crude scoreboard — the evaluations differ wildly in what they measure and in how saturated they are — so the per-benchmark table below matters more than the tally.
| Field | GPT-4 (Mar 2023) | Llama 3.1-405B |
|---|---|---|
| Organization | OpenAI | Meta AI |
| Published | Mar 15, 2023 | Jul 23, 2024 |
| Training compute | 2.1 x 10^25 FLOP | 3.8 x 10^25 FLOP |
| Parameters | 1.8T | 405B |
| Dataset size | 5.4T | 15.6T |
| Training hardware | NVIDIA A100 SXM4 40 GB | NVIDIA H100 SXM5 80GB |
| Chips used | 25,000 | 16,384 |
| Training time | 2.3K h | 2.1K h |
| Training cost (2023 USD) | $37M | $53M |
| Training power draw | 19.9 MW | 22.6 MW |
| Accessibility | API access | Open weights (restricted use) |
| Open weights | No | Yes |
| Country | United States | United States |
| Benchmark | GPT-4 (Mar 2023) | Llama 3.1-405B |
|---|---|---|
| Epoch Capabilities Index(ECI Score) | 125.9 | 129 |
