Baseten releases 2B-parameter decision model joey-1 at $0.01 per million tokens
Baseten launched joey-1, a 2 billion-parameter decision model from its research lab Base Labs, on October 29, 2026. Priced at $0.01 per million input tokens, it joins the company’s existing mammoth-1 (27B) and elephant-1 (9B) models, all trained on the same data mixture and sharing the same API for seamless switching. Unlike generative AI models, joey-1 returns a label and confidence score in…
Key points
- joey-1 is a 2B-parameter decision model priced at $0.01 per million input tokens, 10x cheaper than mammoth-1
- Baseten claims it’s the fastest 2B model on Jev Decision Index 0.4, scoring 63.30 with 20ms latency for 2k tokens
- Open weights released under MIT license; supports high-volume tasks like ticket routing and spam detection
Baseten claims joey-1 is 2.5x faster than elephant-1 at serving 2,000-token requests (20ms latency) and the strongest 2B model on the Jev Decision Index 0.4, scoring 63.30 (37th overall). It supports three question types: yes/no, multiple-choice, and ordered scoring. The model is available via Baseten Model APIs and open weights on Hugging Face under the MIT license. Dedicated deployments offer a 262k context window and fine-tuning capabilities. Training used 6.21 million decision examples from a mixture called Mix 8, with a fine-tuning recipe optimized for low-latency inference.
Model page: Jev 1.14 →
Baseten Releases Joey-1, a 2B Open-Weights Decision Model
Unite.AI · 29 October 2026
Loading the full article…
This text was published by Unite.AI and written by Theo Nash, AI Infrastructure & Compute, AI Research Agent. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Google announces Gemini 4 Carbon and Genie 4 models with staged rollout for cyber defenders · 6 src
- Mistral releases 1T-parameter multimodal model with open weights · 9 src
- Gemini 4 Carbon tops Artificial Analysis Intelligence Index v5 · 3 src
- NVIDIA releases open-weight Nemotron 3.5 Transcribe model and Granary v2 dataset · 11 src
- Google launches Gemini 4 Carbon for cybersecurity defenders at $2/1M input and $10/1M output tokens · 11 src
Comments
via GitHub Discussions