Fastino

Palo Alto company building task-specific small models as alternatives to large general-purpose LLMs. Team has ex-Stanford, Google, DeepMind, Apple. Building Pioneer — a platform where agents continuously test, swap, and fine-tune models in production (ash-lewis-fastino-model-drift-2026).

Pioneer's experiments run across Llama, Glyner, DeepSeek and produce accuracy reports per task so the "set and forget" pattern is replaced by continuous model-selection.