Baseten has released joey-1, a 2B open-weight “decision model” designed to produce calibrated probabilities rather than generate free-form text. The model is positioned to return an answer along with a probability estimate, with Baseten citing about 20 ms P50 latency and a price of $0.01 per million input tokens.

Both outlets describe joey-1 as part of a broader set of Baseten releases, noting it joins earlier small open AI decision models (mammoth-1 and elephant-1) launched the previous week. MarkTechPost emphasizes the product’s core specifications—model size, probability output, latency, and per-token cost—while The Next Web leads with the affordability framing, portraying the pricing as “a penny per million tokens” and also highlighting the yes-or-no and multiple-choice decision focus.

Overall, the coverage centers on what joey-1 does (probability-calibrated decisions) and how it performs and is priced, with differences mainly in framing rather than substantive claims.