Google DeepMind develops advanced AI systems. Its Gemini models, including Gemini 2.5, are multimodal and capable of reasoning through their thoughts before responding, resulting in enhanced performance and improved accuracy.
Bidirectional Encoder Representations from Transformers, 340M-parameter encoder-only model revolutionizing NLP pre-training:contentReference[oaicite:4]{index=4}.
View details →A 70B-parameter model trained on more data to follow the optimal compute/data scaling laws, attaining superior accuracy (e.g., 67.5% on MMLU):contentReference[oaicite:11]{index=11}:contentReference[oaicite:12]{index=12}.
View details →A 280B-parameter Transformer model that was later superseded by the compute-optimal Chinchilla model:contentReference[oaicite:10]{index=10}.
View details →Dialogue-optimized language model (137B parameters) specialized for open-ended conversation generation:contentReference[oaicite:7]{index=7}.
View details →Pathways Language Model, a 540B-parameter Transformer achieving breakthrough performance on many benchmarks:contentReference[oaicite:8]{index=8}.
View details →Second-generation PaLM model (340B parameters) with improved multilingual and reasoning skills, used in Google's Bard chatbot:contentReference[oaicite:9]{index=9}.
View details →Text-to-Text Transfer Transformer (11B parameters) serving as a unified framework for numerous NLP tasks:contentReference[oaicite:6]{index=6}.
View details →Auto-regressive pre-training method (340M parameters) that outperformed BERT on several tasks:contentReference[oaicite:5]{index=5}.
View details →