TL;DR
Paper Instruments has introduced Ultramarine 1 Flash, an open model aimed at finance, law, consulting and operations work. The company reports a score of 59.3 at $0.13 per task on its internal PI Knowledge Work benchmark; independent validation and broader real-world performance are not established in the supplied material.
Paper Instruments has introduced Ultramarine 1 Flash, an open model designed for professional work in finance, law, consulting and operations. The company says it scored 59.3 on its internal PI Knowledge Work benchmark at a reported $0.13 per task, positioning it as a lower-cost option for business tasks such as producing spreadsheets, presentations and documents.
Get art and craft supplies delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
Paper Instruments describes the model as “Fable-class” for knowledge work, a company characterization based on its benchmark results. On PI Knowledge Work, it reports Ultramarine 1 Flash with maximum reasoning at 59.3 points and $0.13 per task. The chart lists Claude Fable 5 at maximum reasoning at 57.8 points and $8.19 per task, GPT-5.6 Sol at maximum reasoning at 55.1 points and $2.09 per task, Claude Opus 5.5 at maximum reasoning at 61.9 points and $6.96 per task, and GPT-6 Sol at maximum reasoning at 61.4 points and $1.26 per task.
The company says the model generates 6.5% fewer output tokens than its base model, GLM-5.3 Flash, across six benchmarks. Its report says the count includes reasoning tokens. For GLM-5.3 Flash, the score shown is public, while token counts came from Paper Instruments’ own runs that it describes as closest to that score.
Paper Instruments lists API prices of $0.15 per million input tokens, $0.03 per million cached input tokens and $0.50 per million output tokens. It says these rates match GLM-5.3 Flash. The per-task costs in its benchmark chart are separate estimates for benchmark runs, rather than the published per-token API rates.
The Cost of Business Tasks
If the reported results hold up in independent evaluations, the model’s low benchmark task cost could interest organizations that run many document and analysis tasks through AI. The benchmark focuses on finished work products, including Excel workbooks, PowerPoint decks and Word documents, as well as tasks inside business systems such as ERP and manufacturing execution systems.
Price alone does not establish whether a model is suitable for consequential business use. Buyers would also need to assess accuracy, reliability, data handling and performance on their own workflows. The supplied report describes a benchmark result, not evidence that the model has been adopted widely or that it outperforms competitors across all professional tasks.
Popular right now
Hisense 55-Inch E6SR Series Hi-QLED 4K UHD Smart Roku TV (55E6SR, 2026 Model)
Vitamix VX1 Professional-Grade Blender, 64 oz, Black
Hisense 43-Inch S5 DécoTV Series 4K UHD Smart Fire TV with Hi-QLED Color, Dolby Vision and Atmos (43S5QF, 2026 Model)
As an affiliate, we earn on qualifying purchases.
How Paper Instruments Tested It
PI Knowledge Work is described by Paper Instruments as an internal benchmark covering finance, law, consulting and operations. The company says its tasks require models to produce finished files or work in live business systems. Task-specific rubrics assess substance, accuracy and presentation, and the company says every model uses the same evaluation harness.
Paper Instruments also reports that PI Knowledge Work scores correlate with the Artificial Analysis Intelligence Index at r = 0.81. That is the company’s reported relationship between the two measures; it does not make the internal benchmark an independent test. The source says API costs use each developer’s official pricing and describes the technical report as containing further details on training data and evaluations.
““An open model for professional work.””
— Paper Instruments
Independent Results Remain Unreported
The supplied material does not identify an independent evaluation of Ultramarine 1 Flash or provide enough methodological detail to reproduce its PI Knowledge Work results. It also does not specify the benchmark’s full task set, sample size, score variability or how the per-task cost estimates were calculated. Those details matter when comparing results across models and settings.
The source does not give an announcement date, licensing terms, model weights or information about data retention and privacy for the API. The label “open model” alone does not clarify which components are publicly available. The model’s performance on particular customer workflows, and the degree to which the benchmark results generalize to routine business use, are also unclear.
Technical Details and Adoption
Paper Instruments points readers to a technical report and its API for further information and access. The company’s published materials do not specify a schedule for additional evaluations or independent testing. For organizations considering the model, the next practical step is to compare it on representative tasks using their own quality, cost and data-handling requirements.
Key Questions
What is Ultramarine 1 Flash?
It is an open model introduced by Paper Instruments for professional tasks in finance, law, consulting and operations. The company describes it as suited to producing documents, spreadsheets and presentations, and to working in business systems.
How did it score in the reported benchmark?
Paper Instruments reports a score of 59.3 at maximum reasoning on its internal PI Knowledge Work benchmark, with an estimated cost of $0.13 per task.
What does API use cost?
The company lists prices of $0.15 per million input tokens, $0.03 per million cached input tokens and $0.50 per million output tokens. These are per-token rates; the reported $0.13 per task is a benchmark cost estimate.
Have independent evaluators confirmed the results?
The supplied source reports Paper Instruments’ own benchmark results and does not identify independent confirmation. It is not yet clear how the model performs across external evaluations or specific customer workflows.
Source: hn
