On June 28th, Google announced the release of Gemma 2 language models for researchers and developers worldwide. It is reported that Gemma 2 has two parameter sizes of 9 billion (9B) and 27 billion (27B). Compared with the first generation, it has higher performance, higher inference efficiency, and significant security improvements built-in. Google claims that the performance of Gemma 2 27B is more competitive than similar products that are twice its size; The performance of 9B is also at the leading level of similar products, superior to Llama 3 8B and other open models.