The reasoning model for the work that matters.
Our flagship reasoning model. Trained on Canadian law, tax, Quebec French, civic values, and graduate-level reasoning and code. Built for contract analysis, policy work, technical problems, and anything that needs careful thinking.
Reasoning · Flagship
Built for thinking it through
The numbers that matter.
Canadian law
45/45
Quebec French
60/60
Canadian values
83.3%
Coding (LiveCodeBench)
77.1%
Three places Ossington 4 earns its keep.
Reasoning, code, and math
Extended thinking mode for contract analysis, policy review, and technical work. Strong on graduate-level science, competitive coding, and AIME-level math — without losing the careful, citation-grounded register Canadian regulated work needs.
Canadian regulated work
Federal and provincial law, CRA and provincial tax, federal policy, and the procedural language Canadian organizations actually use. Familiar with the doctrines, the forms, and where the federal-provincial split matters.
Canadian values + Quebec French
Civic and cultural fluency — Charter values, both official languages as first-class, multicultural framing, and Indigenous context handled with care. Quebec French in standard written register, no phonetic-oral caricature.
Put it to work today, or scope a custom deployment.
Run on Canadian rails.
Post-trained and tuned on a corpus weighted toward Canadian regulated work: federal and provincial law, CRA and provincial tax, Quebec French, federal policy, and Canadian civic and cultural context. We layer in graduate-level science, competition programming, and step-by-step reasoning so the model handles technical work without losing its register on a contract review.
Inference runs on Canadian-managed infrastructure. No US lawful-access surface in the data path. Full architecture documentation available under NDA for procurement reviews and Privacy Impact Assessments.
The full picture, no asterisks.
26B mixture-of-experts foundation. Post-trained by Augure on Canadian-sovereign infrastructure.
Foundation capability
Canadian regulated work · 370-prompt battery
Internal evaluation, May 2026. 370-prompt battery scored by a three-judge ensemble using a five-dimension ordinal rubric with majority-vote pass thresholds. Generalization gap +0.1pt between trained-against and held-out test sets. Full methodology, eval surface, and per-prompt results available under NDA for procurement review.
Put it to work today, or scope a custom deployment.
Ossington 4 across the platform.
Chat
Reasoning-grade chat for everyday regulated work. Document upload, drafting, search.
Learn moreLegal
Contract review with playbook-driven clause analysis. Built on Ossington 4.
Learn moreEnterprise & services
Fine-tune Ossington 4 on your corpus, deploy in your VPC or air-gapped network.
Learn moreTry Ossington 4 free.
Free tier today. Custom deployment tomorrow. Same weights, same Canadian residency, either way.