Fine-tuned 2B models match frontier API classification quality at 1% cost

Airbnb and Meta engineering teams showed that fine-tuned 2B open-source models can match frontier API quality on production classification tasks, achieving F1 of 0.902 with 35ms median latency at roughly 1% of the cost.

2026-11-05 ~ 2026-11-05 · 2 related posts