Baseten releases Joey‑1, a 20 ms decision model

What is changing?

Baseten’s research team, Base Labs, released a new decision model called Joey‑1 on October 29. According to the provider, the model returns probabilities and confidence scores for yes/no, multiple-choice and ordered-scoring questions in a single pass rather than generating free-form text.

Joey‑1 is post-trained from Qwen3.5‑2B and supports text and image inputs. Baseten reports a P50 end-to-end latency of 20 ms for an average 2,000-token request, roughly 2.5 times faster than Elephant‑1 at 49 ms. These are provider measurements, and the detailed testing methodology has not yet been published.

The model is available through Baseten Model APIs at $0.01 per million input tokens, while its open weights are published on Hugging Face under the MIT licence. The API version offers a 16K context window and dedicated deployments support 262K. It may reduce cost for high-volume classification work such as support routing, spam checks and policy scoring, but teams should measure thresholds, calibration and error distributions on their own data before using it for consequential decisions.