Inflection-2 (Inflection AI) and U-PaLM (540B) (Google) are both frontier AI models. Inflection-2 was published in November 2023 and U-PaLM (540B) in October 2022.
Inflection-2 was trained on 10^25 FLOP, about 4.0x the compute of U-PaLM (540B) at 2.5 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.
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 | Inflection-2 | U-PaLM (540B) |
|---|---|---|
| Organization | Inflection AI | |
| Published | Nov 22, 2023 | Oct 20, 2022 |
| Training compute | 10^25 FLOP | 2.5 x 10^24 FLOP |
| Parameters | — | 540B |
| Dataset size | — | 1.3B |
| Training hardware | NVIDIA H100 SXM5 80GB | Google TPU v4 |
| Chips used | 5,000 | 512 |
| Training time | — | 120 h |
| Training cost (2023 USD) | $13M | — |
| Training power draw | 6.9 MW | 348.3 kW |
| Accessibility | Hosted access (no API) | Unreleased |
| Open weights | No | No |
| Country | United States | United States |
