Decisio newshttps://opendecisio.com/news/Releases, results and announcements from the Decisio project.enThe tuned 12B can think on hard choice questionshttps://opendecisio.com/news/tuned-12b-commit-think/https://opendecisio.com/news/tuned-12b-commit-think/Mon, 26 Oct 2026 00:00:00 GMTA third setting for the tuned 12B, commit-think, reads a few unsure choice questions again after a short reasoning step, and is right more often on hard decisions.A tuned Gemma 4 12B: right more often on hard, rule-heavy decisionshttps://opendecisio.com/news/tuned-12b/https://opendecisio.com/news/tuned-12b/Sun, 25 Oct 2026 00:00:00 GMTOur tuned 12B is right about 7 points more often than the untrained 12B in the same layout on held-out hard decisions, and about 3 points more on everyday wording. Every way to run it.decisio 0.27.0: truncation fixed under the header layout, and opt-in flavorshttps://opendecisio.com/news/decisio-0-27-0/https://opendecisio.com/news/decisio-0-27-0/Sat, 24 Oct 2026 00:00:00 GMT0.27.0 fixes --over-limit truncate under the header layout, and adds flavors, reading short requests twice and think temperatures, all off unless you ask for them.decisio 0.26.0: the Gemma models read your questions first, and answers changehttps://opendecisio.com/news/decisio-0-26-0/https://opendecisio.com/news/decisio-0-26-0/Fri, 23 Oct 2026 00:00:00 GMT0.26.0 lists a request's questions before its text on the Gemma models, which answered more accurately on the 31B; anything taught keeps the old layout, and --layout plain restores it.decisio 0.24.0: the 12B on a 32 GB card, batched serving, and thinking that reasonshttps://opendecisio.com/news/decisio-0-24-0/https://opendecisio.com/news/decisio-0-24-0/Thu, 22 Oct 2026 00:00:00 GMT0.24.0 adds the Gemma 4 12B in FP8 (full 65,536 tokens on a 32 GB card), faster opt-in batched serving, and a fix so think-when-unsure reasons.decisio 0.25.0: taught tasks survive restarts, and replicas share themhttps://opendecisio.com/news/decisio-0-25-0/https://opendecisio.com/news/decisio-0-25-0/Thu, 22 Oct 2026 00:00:00 GMT0.25.0 saves taught tasks to disk so a restart keeps them, and shares them between replicas through object storage, so a task taught once reaches every server.decisio 0.15.0: the 31B on a 48 GB Machttps://opendecisio.com/news/decisio-0-15-0/https://opendecisio.com/news/decisio-0-15-0/Sat, 17 Oct 2026 00:00:00 GMT0.15.0 adds an opt-in 4-bit build of the Gemma 4 31B for Macs with 48 GB or more. It costs 1.5 points of MMLU-Pro and is slow on long documents.decisio 0.13.0: long documents, read wholehttps://opendecisio.com/news/decisio-0-13-0/https://opendecisio.com/news/decisio-0-13-0/Thu, 15 Oct 2026 00:00:00 GMT0.13.0 can read documents of about 98,000 tokens on the Gemma 4 31B and 65,536 on the 12B, as an opt-in. It is off by default, and the first question is slow.decisio 0.11.1: the 12B no longer slows down when new documents arrive back to backhttps://opendecisio.com/news/decisio-0-11-1-12b-back-to-back/https://opendecisio.com/news/decisio-0-11-1-12b-back-to-back/Wed, 14 Oct 2026 00:00:00 GMTTwo server changes: the Gemma 4 12B now registers a document's boundary before the question, and every request records how long it waited for the engine.decisio 0.12.0: certified answershttps://opendecisio.com/news/decisio-0-12-0/https://opendecisio.com/news/decisio-0-12-0/Wed, 14 Oct 2026 00:00:00 GMT0.12.0 lets you certify which answers to act on, from your own labelled examples. It also refuses non-JSON POSTs and sets memory limits on a Mac.Fifty questions about one document, five systems, one boxhttps://opendecisio.com/news/same-box-speed-50-questions/https://opendecisio.com/news/same-box-speed-50-questions/Wed, 14 Oct 2026 00:00:00 GMTHow long five systems take to answer 50 questions about one document, measured in one session on one RTX PRO 6000. Time, not accuracy.