Ultramarine 1 Flash

Ultramarine 1 Flash is an open, production-focused large language model from Paper Instruments built for professional knowledge work in finance, law, consulting, and operations. It delivers Fable-class capability at a dramatically lower price point - advertised as roughly 60× cheaper than Fable - scoring strongly on multiple benchmarks (JobBench 62.9, PI Knowledge Work 59.3) while hitting a max reasoning score of 59.3 at $0.13 per task. Compared with larger proprietary models, it occupies a cost-quality frontier: it outperforms some pricier models at a fraction of their cost and approaches the performance of premium offerings like Claude Opus and GPT-6 variants when normalized for price. API pricing matches GLM-5.3 Flash per-token rates, and model weights are forthcoming.

Benchmarks show practical strengths: Ultramarine produces 6.5% fewer output tokens than GLM-5.3 Flash across six professional benchmarks, indicating more efficient reasoning paths. PI Knowledge Work tasks require finished Excel workbooks, PowerPoint decks, Word documents, and interactions with enterprise systems, scored for substance, accuracy, and presentation; Ultramarine is tuned to produce clear prose and strong visual hierarchy suitable for client deliverables. Example outputs include reconciled multi-period financial slides (Nucor) with accurate subtotals and a clear narrative of earnings and sales mix changes, demonstrating readiness for analyst workflows.