Gemini 1.0 Ultra (Google DeepMind) and GPT-4 (Mar 2023) (OpenAI) are both frontier AI models. Gemini 1.0 Ultra was published in December 2023 and GPT-4 (Mar 2023) in March 2023.
Gemini 1.0 Ultra was trained on 5 x 10^25 FLOP, about 2.4x 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.
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 | Gemini 1.0 Ultra | GPT-4 (Mar 2023) |
|---|---|---|
| Organization | Google DeepMind | OpenAI |
| Published | Dec 6, 2023 | Mar 15, 2023 |
| Training compute | 5 x 10^25 FLOP | 2.1 x 10^25 FLOP |
| Parameters | — | 1.8T |
| Dataset size | — | 5.4T |
| Training hardware | Google TPU v4 | NVIDIA A100 SXM4 40 GB |
| Chips used | 57,000 | 25,000 |
| Training time | 2.4K h | 2.3K h |
| Training cost (2023 USD) | $31M | $37M |
| Training power draw | 38.4 MW | 19.9 MW |
| Accessibility | API access | API access |
| Open weights | No | No |
| Country | United States | United States |
