Schematron V2 Turbo

Availability: Served on Inference.net's own OpenAI-compatible API at api.inference.net/v1; the served build is tagged inference-net/schematron-v2-turbo-20260902. Two caveats a reader may hit. First, the context window: the catalogue and the model page both display "125K tokens", while the record behind them says maxContextSize 128000. These are the same number — the site renders every window as floor(tokens/1024)K, which is why 1,000,000 shows as "977K" and 200,000 as "195K" on the same page — so the published 128,000 is the maker's figure, not a third-party one. Second, the release date: the catalogue record carries releaseDate 2026-04-16, which cannot be this SKU's launch (the served build is dated 2026-09-02, the SKU first appears in the OpenRouter catalogue on 2026-09-12, and the field repeats claude-opus-4-7's date on the same page), so we publish the first date on which it is provably purchasable rather than the maker's field. Schematron V1 (Schematron-3B, Schematron-8B) has public weights on HuggingFace; V2 does not — inference-net/schematron-v2-granite-4.0-h-micro is not in the org's public model list and answers 401 — so this row is open_source false.