Claude 2 (Anthropic) and GPT-4 (Mar 2023) (OpenAI) are both frontier AI models. Claude 2 was published in July 2023 and GPT-4 (Mar 2023) in March 2023.
GPT-4 (Mar 2023) was trained on 2.1 x 10^25 FLOP, about 5.4x the compute of Claude 2 at 3.9 x 10^24 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.
GPT-4 (Mar 2023) 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 | Claude 2 | GPT-4 (Mar 2023) |
|---|---|---|
| Organization | Anthropic | OpenAI |
| Published | Jul 11, 2023 | Mar 15, 2023 |
| Training compute | 3.9 x 10^24 FLOP | 2.1 x 10^25 FLOP |
| Parameters | — | 1.8T |
| Dataset size | — | 5.4T |
| Training hardware | — | NVIDIA A100 SXM4 40 GB |
| Chips used | — | 25,000 |
| Training time | — | 2.3K h |
| Training cost (2023 USD) | $5M | $37M |
| Training power draw | — | 19.9 MW |
| Accessibility | API access | API access |
| Open weights | No | No |
| Country | United States | United States |
| Benchmark | Claude 2 | GPT-4 (Mar 2023) |
|---|---|---|
| Epoch Capabilities Index(ECI Score) | 119.6 | 125.9 |
