Baseten Bison-1 and Groundhog-1 Top Decision Model Index

Baseten has launched two new open-weight decision models, bison-1 and groundhog-1, both built on the Gemma 4 architecture. The release positions bison-1 at the top of the AA Decision Index v2.1 with a score of 76, marking a significant milestone in the rapidly evolving decision model category. Both models support 128k context windows and multimodal input capabilities including image processing.
Model Architecture and Capabilities
Bison-1 and groundhog-1 leverage the Gemma 4 foundation to deliver specialized decision-making capabilities. The models process both text and image inputs, enabling applications that require visual understanding alongside reasoning tasks. The 128k context window allows for extended document analysis and complex multi-step reasoning scenarios that require substantial information retention.
Both models are released as open-weight implementations, giving developers access to the underlying parameters without proprietary restrictions. This approach aligns with the growing demand for transparent AI systems that can be deployed in controlled environments where data governance and model auditability are critical.
Benchmark Performance
Bison-1's score of 76 on the AA Decision Index v2.1 places it ahead of competing decision models in the benchmark rankings. The index evaluates models on their ability to make sound judgments across diverse scenarios, from business logic to ethical reasoning tasks. Groundhog-1, while not explicitly scored in the initial release, shares the same architectural foundation and training methodology.
The decision model category has seen increased competition in 2026, with several labs releasing specialized systems designed for agentic workflows and autonomous decision-making. Baseten's entry demonstrates the company's focus on practical enterprise applications rather than general-purpose language tasks.
Release Date and Availability
The models were officially released on November 20, 2026. Baseten has made both bison-1 and groundhog-1 available through its deployment platform, targeting organizations building AI-powered decision systems. The open-weight nature of the release allows for self-hosting and fine-tuning, addressing privacy concerns in regulated industries.
What This Means
Baseten's dual model launch underscores the specialization trend in AI development, where purpose-built models outperform general systems in domain-specific tasks. The strong benchmark performance suggests decision models may become a distinct category alongside conversational and embedding models. For enterprises evaluating AI infrastructure, the availability of high-performing open-weight alternatives provides leverage in negotiations with closed-source providers while maintaining control over sensitive decision-making processes.
on Emergent today





