Baseten has launched two new open-weight decision models, Bison-1 and Groundhog-1, both built upon Google's Gemma 4 architecture. These models are designed to provide calibrated probabilities for yes/no, choice, and score-based questions, rather than generating text. Bison-1, a 31B dense model, achieved a score of 76 on the Artificial Analysis Decision Index v2.1, outperforming its predecessor Mammoth-1 and Liquid AI's d1.2-ultra. Both new models support an extended context window of up to 128k tokens and can process image inputs, enhancing their utility for complex routing, triage, and guardrail applications. AI
IMPACT These models offer improved decision-making capabilities with extended context, potentially enhancing applications in routing, triage, and guardrail systems.
RANK_REASON New model release from a company that is not a Tier-1 frontier lab, but the models are built on a Tier-1 model (Gemma 4) and achieve top rankings on a specialized benchmark. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- Artificial Analysis Decision Index v2.1
- Baseten
- d1.2-ultra
- Gemma 4
- groundhog-1
- Hugging Face
- Liquid AI
- Mammoth-1
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →