Baseten has introduced joey-1, a 2 billion-parameter decision model developed by its internal research unit, Base Labs. Unlike generative models, joey-1 does not produce text responses. It returns labels and calibrated probabilities, targeting high-throughput routing, classification, and safety guardrail checks. The company priced the model at $0.01 per million input tokens and released its weights on Hugging Face under the MIT license.
joey-1 is the third model in Baseten’s decision-model lineup, following mammoth-1 with 27 billion parameters and elephant-1 with 9 billion parameters, both released last week. Baseten said the new model was post-trained on Qwen3.5-2B and can take either text or image state inputs. It is designed to answer three types of questions: multiple choice, yes-or-no, and scoring tasks.
Baseten reported a P50 end-to-end latency of 20 milliseconds. On the public partition of the Jev Decision Index 0.4 leaderboard, joey-1 posted a score of 63.30, ranking 37th out of 230 models. It was also the top-scoring model among systems with fewer than 4 billion parameters, according to MarkTechPost.
Baseten has released joey-1, a 2 billion-parameter decision model built by its internal research lab, Base Labs. The model does not generate text. Instead, it returns labels and calibrated probabilities for high-throughput routing, classification, and safety guardrail checks.
Baseten priced joey-1 at $0.01 per million input tokens. Its model weights have been open-sourced on Hugging Face under the MIT license.
A new addition to Baseten’s decision-model family
joey-1 is the third member of Baseten’s decision-model family. The first two, mammoth-1 and elephant-1, were released last week with 27 billion and 9 billion parameters, respectively.
Post-trained on Qwen3.5-2B
According to Baseten, joey-1 was post-trained on Qwen3.5-2B. It accepts either text or image state inputs and is built to answer three categories of questions: multiple choice, yes-or-no, and scoring.
Baseten reported a P50 end-to-end latency of 20 milliseconds for the model.
Leaderboard result
On the public partition of the Jev Decision Index 0.4 leaderboard, joey-1 scored 63.30 and ranked 37th out of 230 models. It was also the highest-scoring model among those with fewer than 4 billion parameters.
The development was reported by MarkTechPost.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
