Joey-1

Joey-1 (joey-1) is a 2B open-weights decision model from Base Labs (Baseten’s research lab), released October 29, 2026. It is the fastest/cheapest tier in the family alongside mammoth-1 (27B) and elephant-1 (9B), sharing one decision API.

Library: baseten.co/library/joey-1 · Blog: Decisions for a penny · OpenRouter: baseten/joey-1

Key points

1. Decide, don’t generate: given a state plus typed questions (choice / yes-no / ordered scoring), one forward pass returns labels with calibrated probabilities—no free-form text to parse.
2. Speed & price: vendor reports ~20 ms P50 e2e for a typical 2k-token request (~2.5× faster than elephant-1) at $0.01 / M input tokens (~1/10 of mammoth-1).
3. Base: post-trained from Qwen3.5-2B; text+image input; 16k context on Model APIs, up to 262k on dedicated deployments.
4. License & access: MIT open weights on Hugging Face; API /v1/systemone with model id decision-models/baseten-joey-2b-latest.
5. Use cases: ticket routing, spam/phishing flags, guardrails/jailbreak checks, small↔frontier model routing.

Why it matters

Official blog + new OpenRouter listing + X launch posts in this window complete the searchable Base Labs decision-model family next to Mammoth-1 / Elephant-1.

How to try it

1. Read the launch blog and model library page.
2. Call Baseten Model APIs /v1/systemone with decision-models/baseten-joey-2b-latest, or pick baseten/joey-1 on OpenRouter.
3. Self-host via MIT open weights on Hugging Face.

Prefer Joey-1 for high-volume / latency-sensitive routing and guardrails; escalate uncertain cases to elephant-1 or mammoth-1.