Gemma 7B (initial pretrained)
The initial Gemma 7B pretrained model was announced with conditional open weights on February 21, 2024. Its public model card distinguishes it from the instruction-tuned variant and reports an 8192-token training context. No hosted API date or immutable checkpoint is assigned.
Benchmark scores
Access
Historical conditional open-weight release under Gemma terms, not Apache 2.0. The public card identifies the named base variant, but model files require licence acceptance and were not accessed. The 8192-token figure is a reported training context, not a guarantee for a hosted API. Current availability and retirement are unassessed.