Turbopuffer Partners with Applied Compute on Code Search Model

Turbopuffer announced a partnership with Applied Compute to post-train a small model for large-scale code search. The official post states the model operates over precomputed indexes across 300 repositories. It claims results roughly 3x faster than filesystem and grep while cutting marginal search costs by up to 100x compared with frontier models. The linked turbopuffer blog post describes the work as showing that a specialized open-weight model can deliver strong search quality at lower latency and expense than general-purpose approaches.