← All models

RecurrentGemma 9B

Google

RecurrentGemma 9B is Google DeepMind's larger open-weight, English text-to-text pre-trained model built on the Griffin architecture, which mixes gated linear recurrences with 2,048-token local attention so its inference state stays fixed in size. Google's technical report gives 8.58 billion total parameters (7.53 billion non-embedding), while the Hugging Face safetensors checkpoint stores 9,628,553,216 values, so total_parameters is left unset; it was trained on 2 trillion tokens of 8,192-token sequences.

Benchmark scores