groundhog-1
Intro
groundhog-1 is a decision model that Baseten released on 20 November 2026 alongside bison-1, also built on Google’s Gemma 4. Official blog: Meet bison-1 and groundhog-1. The official account @baseten introduced both models the same day: improved quality versus mammoth-1 and elephant-1, similar speed, a 128k context window, and image-and-text input.
The official post places groundhog-1 in the smaller slot of the family, next to elephant-1: it is priced like elephant-1, and the lab says “we expect groundhog-1 to match the performance of mammoth-1 on the official AA index.” It is live on Baseten Model APIs, with open weights on Hugging Face. For the current #1 ranking, use the same-day bison-1 result (76 on AA Decision Index v2.1); do not treat that expected comparison as a published groundhog-1 score.
Advertisement
Explain
What problem it targets
Like bison-1, groundhog-1 is meant to offer a usable decision model on a Gemma 4 backbone, with eight times the previous context and multimodal input. It sits in the family slot that corresponds to elephant-1: the same American base and API, at a smaller price and size point.
Official differences from bison-1
- Both have a 128k context window, text-and-image input, and the same Model APIs;
- bison-1 has a published AA Decision Index v2.1 score of 76 (first of 19) and a Brier score of 0.14;
- groundhog-1 is described as expected to match mammoth-1 on the official AA index, and is priced like elephant-1;
- The blog says the choice can come down to the decision at hand, because the prices already line up with mammoth-1 and elephant-1.
Limits
Until an official index score is published, “expected to match mammoth-1” is not a recorded leaderboard result. Exact open-weight repository names should be taken from Baseten or Hugging Face at the time you use them.
Usage
Who it is for
- Teams that want decision-making on a Gemma 4 backbone at a price point closer to elephant-1;
- Users of elephant-1 who want to try an American base, 128k context, and image input.
How to use it
- Read the official blog: https://www.baseten.co/blog/meet-bison-1-and-groundhog-1 ;
- On Baseten Model APIs, change the model name to
groundhog-1; - For weights, follow the official pointer to Hugging Face;
- When you compare quality, keep bison-1’s published AA score separate from groundhog-1’s expected match to mammoth-1.