Google has unveiled its latest large language model, Gemma 3, which asserts near parity to DeepSeek AI's R1 model in accuracy while utilizing only a single H100 GPU. The model reportedly achieves an Elo score of 1338 compared to R1's 1363, claiming a significant gain in computational efficiency. Google emphasizes this balance as a crucial innovation, highlighting that Gemma 3 outperforms several competing models. With the ability to run on just one GPU or TPU, Gemma 3 aims to provide developers with scalable and engaging AI capabilities.
Google claims that Gemma 3 achieves 98% of the accuracy score of DeepSeek's R1 while utilizing just one Nvidia H100 GPU, illustrating significant efficiency.
Gemma 3 is positioned as the most capable AI model to run on a single GPU or TPU, indicating a substantial leap in resource efficiency.
Collection
[
|
...
]