Baseten announced two new decision models, bison-1 and groundhog-1, built on Google’s Gemma 4 backbone. Both support text and image input and have a 128k token context window, eight times larger than Baseten’s prior models. Bison-1 tops the Artificial Analysis Decision Index with a score of 76, while groundhog-1 is priced like the earlier elephant-1 model. The models are live on Baseten’s Model APIs and their weights are open on Hugging Face.
Why it matters
Teams that need long context for accurate decisions can use these models now without extra summarization steps.