Study: embeddings plus simple classifiers beat Jev model

A Voyage AI employee reported that using an embedding model plus a simple sklearn classifier was cheaper and significantly more accurate than the Jev decision model when labeled data is available, cutting error by 37%. The author noted the gain comes from training labels that Jev does not use, so zero-shot cases may differ.

• Developer launches picture quiz game built on Jev vision model2026-10-20
• Cloudflare previews Clairic extraction model2026-10-19
• Developer releases open-source model Herbert, seeks feedback2026-10-19
• Vercel data: Jev is fastest-adopted AI model ever2026-10-16
• Decision 3.0 open multimodal decision models released2026-10-12