Decisions for a penny: introducing joey-1, our fastest decision model

Baseten has released joey-1, a 2-billion-parameter decision model designed for high-volume, latency-sensitive classification and approval tasks that require a label and calibrated confidence score rather than generated text. The company says joey-1 achieves a median end-to-end latency of 20 milliseconds for typical 2,000-token requests, costs $0.01 per million tokens, and is its strongest decision model with 3 billion or fewer parameters in internal testing. It joins the 9B elephant-1 and 27B mammoth-1 models in Base Labs’ decision-model family, which shares training data, API compatibility, text-and-image support, and a tiered approach in which joey-1 handles inexpensive first-line decisions while larger models serve as escalation options. Post-trained from Qwen3.5-2B, joey-1 is available through Baseten Model APIs, dedicated deployments, and open weights on Hugging Face; API access supports a 16,000-token context window, while dedicated deployments offer up to 262,000 tokens, workload-specific optimization, and fine-tuning options.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.