Gemma 4 26B TL5
Gemma 4 26B A4B is Google DeepMind's open-weight, instruction-tuned multimodal Mixture-of-Experts model, built for advanced reasoning and efficient inference. It combines approximately 26B total parameters with 4B active per token and supports a 256K-token context window. The model accepts text and images, including video represented as sequences of frames, and generates text; it does not support audio input. Native function calling and configurable thinking enable multi-step agents, software engineering, and tool-driven workflows. Vision capabilities cover document and PDF parsing, chart interpretation, UI understanding, and multilingual OCR. Trained on data covering more than 140 languages, Gemma 4 also supports multilingual question answering and long-document analysis. The weights are released under Apache 2.0 for commercial use.
Gemma 4 26B TL5
Gemma 4 26B A4B is Google DeepMind's open-weight, instruction-tuned multimodal Mixture-of-Experts model, built for advanced reasoning and efficient inference. It combines approximately 26B total parameters with 4B active per token and supports a 256K-token context window. The model accepts text and images, including video represented as sequences of frames, and generates text; it does not support audio input. Native function calling and configurable thinking enable multi-step agents, software engineering, and tool-driven workflows. Vision capabilities cover document and PDF parsing, chart interpretation, UI understanding, and multilingual OCR. Trained on data covering more than 140 languages, Gemma 4 also supports multilingual question answering and long-document analysis. The weights are released under Apache 2.0 for commercial use.