TL;DR
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 has introduced Ultramarine 1 Flash, an open model aimed at finance, law, consulting and operations work. The company reports a score of 59.3 on its internal PI Knowledge Work benchmark at an estimated $0.13 per task; the results and cost comparisons come from the company’s evaluation, not an independent audit described in the report.
Paper Instruments has introduced Ultramarine 1 Flash, an open model designed for professional tasks in finance, law, consulting and operations. The company says the model scored 59.3 on its internal PI Knowledge Work benchmark at an estimated $0.13 per task, putting its reported cost below several higher-scoring or similarly scoring systems in the comparison.
The company says Ultramarine 1 Flash is built on GLM-5.3 Flash and uses the same listed API rates: $0.15 per million input tokens, $0.03 per million cached input tokens and $0.50 per million output tokens. Paper Instruments reports that across six benchmarks the model generated 6.5% fewer tokens than its base model, including reasoning tokens. The report does not give the six benchmark names in the supplied material.
On PI Knowledge Work, the company reports a score of 59.3 at $0.13 per task for Ultramarine 1 Flash with maximum reasoning. Its chart lists Claude Fable 5 at maximum reasoning as scoring 57.8 at $8.19 per task, and GPT-5.6 Sol at maximum reasoning as scoring 55.1 at $2.09. Claude Opus 5.5 at maximum reasoning is listed at 61.9 and $6.96 per task, while GPT-6 Sol at maximum reasoning is listed at 61.4 and $1.26.
Paper Instruments describes the model as “Fable-class” for knowledge work and says the benchmark’s scores correlate with the Artificial Analysis Intelligence Index at r = 0.81. These are the company’s characterizations and figures. The source material does not provide an independent assessment of the benchmark, its scoring, or the reported task costs.
A Lower Reported Cost Per Task
If the reported benchmark and cost estimates hold up in independent use, Ultramarine 1 Flash could make capable models more affordable for organizations that run many document, spreadsheet or business-system tasks. The company’s stated $0.13 per task is far below the listed costs for several models with nearby scores on its benchmark, though comparisons depend on the tasks, model settings and cost assumptions used.
The release also highlights a distinction between API token prices and full task costs. Token rates describe charges per volume of input or output. The benchmark’s per-task estimate additionally reflects how much work the model uses to complete the tested task. Organizations weighing a model would need to consider quality, reliability, speed, data handling and integration costs alongside the quoted task estimate.
Paper Instruments says the model produces finished Excel workbooks, PowerPoint decks and Word documents, and can work inside live business systems such as ERP and MES. Those capabilities could matter to teams evaluating models for routine professional workflows, but the source gives no deployment case studies or independent customer results.
professional AI task automation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How the Company Tested It
PI Knowledge Work is described by Paper Instruments as an internal benchmark for professional work across finance, law, consulting and operations. The tasks require models to create finished office documents and operate in business software. According to the company, each task has its own rubric covering substance, accuracy and presentation, and models run through the same harness.
The report says benchmark costs use each developer’s official API pricing. It also notes a qualification for GLM-5.3 Flash: its score is public, while its token usage comes from Paper Instruments’ own runs that it says were closest to that score. This means the displayed score and cost inputs do not all have the same source. The report points readers to a technical report for more information about the professional data and evaluations used to train the model.
The company presents the results as evidence of a cost-quality position for Ultramarine 1 Flash. The material supplied here does not state the benchmark’s release date, disclose task-level results, or show how outside evaluators reproduced the comparisons.
““At $0.13 per task””
— Paper Instruments, in its Ultramarine 1 Flash report
business document automation tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Results Are Not Shown
The supplied report does not identify an independent evaluator or provide enough task-level data to reproduce the 59.3 score and $0.13 estimate. It is also unclear how representative the benchmark tasks are of the work organizations handle in practice, and whether results would hold across different prompts, software systems or data.
The source does not give a publication date, model license terms, availability details, latency figures, or information about privacy and data retention for API use. Its “Fable-class” description is the company’s own comparison, and the report does not define that label as an external standard.
AI-powered Excel and PowerPoint generators
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Technical Details and Availability
Paper Instruments directs readers to a technical report for further information about its training data and evaluations, and lists an Ultramarine API as the access route. The supplied source does not specify when the model became available, the terms of its open-model release, or whether more benchmark data will be published.
For prospective users, the next useful evidence would include the full task set, scoring rubrics, reproducible cost calculations and tests by independent teams. Until those details are available, the published numbers are best read as Paper Instruments’ reported results, rather than a guarantee of performance or cost for every workplace deployment.
enterprise AI integration solutions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Ultramarine 1 Flash?
It is an open model from Paper Instruments intended for professional work in finance, law, consulting and operations, according to the company.
How did it score?
Paper Instruments reports a score of 59.3 with maximum reasoning on its internal PI Knowledge Work benchmark. The supplied material does not include independent verification.
What does the $0.13 per task figure mean?
It is the company’s estimated cost per task on PI Knowledge Work. The report also lists API rates of $0.15 per million input tokens, $0.03 per million cached input tokens and $0.50 per million output tokens.
Was the benchmark independently reviewed?
The supplied report does not identify an independent review. Paper Instruments says the benchmark uses task-specific rubrics and a shared harness, but outside replication details are not provided here.
Where can users find more information?
Paper Instruments points readers to its technical report and lists an Ultramarine API. The supplied material does not state the release date or license terms.
Source: hn
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
