Baseten Releases Joey-1: 2B Open-Weights Decision Model
Baseten released joey-1, a 2B open-weights decision model developed by research lab Base Labs and post-trained from Qwen3.5-2B. Baseten reports 20 ms P50 end-to
Image source: https://www.baseten.co/blog/decisions-for-a-penny-introducing-joey-1/
Source: https://www.baseten.co/blog/decisions-for-a-penny-introducing-joey-1/
Related source: https://www.baseten.co/blog/decisions-for-a-penny-introducing-joey-1/
Baseten released joey-1 on October 29, 2026, a 2B-parameter decision model developed by its research lab Base Labs and post-trained from Qwen3.5-2B. Its open weights are available under the MIT license.
Unlike generative LLMs, joey-1 does not generate text. It returns a label and a confidence score for that label in one forward pass.
Key Features and Benchmarks
- Architecture and Training: Post-trained from Qwen3.5-2B using 6.21 million decision examples (Mix 8) with the vision encoder frozen and a maximum sequence length of 16,384 tokens.
- Benchmark Performance: Scored 63.30 on the Jev Decision Index 0.4 public split, placing 37th of 230 models and recording the highest score among models under 4B parameters.
- Latency: Serves an average 2,000-token request at a P50 end-to-end latency of 20 ms in production scale.
API Pricing and Availability
joey-1 is available on Baseten Model APIs under the model ID decision-models/baseten-joey-2b-latest and as open weights on Hugging Face.
- Pricing: $0.01 per million input tokens, with cached input and output listed at $0.00.
- Use Cases: Baseten lists ticket triage and routing, spam and phishing detection, prompt guardrails, and routing requests between small and frontier LLMs.