Ultramarine 1 Flash is an open, production-oriented large language model optimized for professional knowledge work across finance, law, consulting, and operations. Benchmarked on multiple suites (JobBench, PI Knowledge Work, FrontierFinance, AutomationBench, DiligenceBench, APEX-Agents), it posts top or near-top scores (e.g., JobBench 62.9, PI Knowledge Work 59.3) while running at a fraction of competitor cost: about $0.13 per task, roughly 60x cheaper than high-end Fable-class offerings and outperforming Claude Fable 5 and GPT-5.6 Sol on cost-quality tradeoffs. Compared against GLM-5.3 Flash, Ultramarine generates 6.5% fewer output tokens (30.5M vs 32.7M across the six benchmarks), improving token efficiency while sharing the same per-million-token API pricing.
Design choices emphasize client-readiness: tasks include delivering finished Excel workbooks, PowerPoint decks, and Word documents and interacting with live business systems (ERP/MES), with rubric scoring for substance, accuracy, and presentation. Prose is described as clean and hierarchically organized to match analyst standards; an example slide reconstructs multi-period Nucor segment sales and reconciled subtotals. Model weights are forthcoming, and an API, technical report, and code links are provided for integration and evaluation.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.