Ollama halves time-to-first-token by caching resolved model metadata

Ollama v0.32.15, published 19 August 2026, caches resolved model metadata between requests, which the project measured as cutting time-to-first-token roughly in half — from about 995 milliseconds to about 524 in its own benchmarks. The same release adds a desktop onboarding flow on first launch and fixes a case where chat and generate could wedge after a mid-stream parser error.

What it means Time-to-first-token is what a local model feels like, so halving it changes whether self-hosting is pleasant enough to use for real work.