Baseten made two open‑weight decision models—bison-1 and groundhog-1—available via its Model APIs on November 20, 2026. Both run on Google’s Gemma 4 backbone and accept text and image inputs.
Record performance on the AA Decision Index
On the Artificial Analysis Decision Index v2.1, bison‑1 earned a score of 76, ranking #1 among 19 decision models. The next‑best model, d1.2‑ultra, scored 75, while Baseten’s earlier mammoth‑1 posted 71.
Long context and multimodal support
Both models provide a 128 k‑token context window, eight times larger than Baseten’s previous offerings. This lets teams feed full policies, invoices or conversation histories without summarising first.
Pricing and access
Baseten charges $0.10 per million tokens for bison‑1 and $0.035 per million tokens for groundhog‑1. The weights are published openly on Hugging Face, and the same API used for earlier models applies.
Why it matters
Teams that need high‑accuracy decisions over long documents can now use an American‑backed model without building their own infrastructure, paying only a few cents per million tokens.
What it means for you
A Baseten customer on the standard plan can start using bison‑1 today by swapping the model name in their API call, paying $0.10 per million tokens for the higher‑accuracy, long‑context option.
Our take: Baseten’s new American‑backed models raise the accuracy ceiling for decision‑making while keeping costs modest, but only early adopters can verify real‑world speed and calibration claims.