TL;DR
Get audio and creator gear 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 based on GLM-5.3 Flash and aimed at finance, law, consulting and operations work. On the company’s internal PI Knowledge Work benchmark, it scored 59.3 at an estimated $0.13 per task; those results and cost estimates come from the company, and have not been independently verified in the supplied material.
Paper Instruments has introduced Ultramarine 1 Flash, an open model designed for professional tasks in finance, law, consulting and operations, and says it scored 59.3 on its internal PI Knowledge Work benchmark at an estimated $0.13 per task. The figures are the company’s own evaluation, rather than an independent assessment, but frame the release as a bid to make document, spreadsheet and business-system work cheaper to automate.
Paper Instruments says Ultramarine 1 Flash is based on GLM-5.3 Flash and is priced on the Ultramarine API at $0.15 per million input tokens, $0.03 per million cached input tokens and $0.50 per million output tokens. The company says those rates match GLM-5.3 Flash. It also reports that the new model generated 6.5% fewer tokens across six benchmarks than its base model. The report counts output tokens, including reasoning; for the GLM comparison, the company says it used its own runs closest to the public score.
In the company’s comparison table, Ultramarine 1 Flash with maximum reasoning scores 59.3 at $0.13 per task. Paper Instruments says that is above Claude Fable 5 at maximum reasoning, listed at 57.8 and $8.19 per task, and GPT-5.6 Sol at maximum reasoning, listed at 55.1 and $2.09. The report places the model within three score points of Claude Opus 5.5 at maximum reasoning, which scores 61.9 at $6.96, and GPT-6 Sol at maximum reasoning, which scores 61.4 at $1.26.
The benchmark is described as an internal suite of tasks requiring finished Excel workbooks, PowerPoint decks and Word documents, as well as work in live business systems such as ERP and manufacturing execution systems. Paper Instruments says task-specific rubrics assess substance, accuracy and presentation, and that models use the same harness. It says task costs use each developer’s official API pricing. The supplied report does not provide independent replication of those procedures or scores.
A Low-Cost Bid for Office Tasks
If the results hold up in independent testing and real deployments, the claimed price-performance could matter to organizations that run large numbers of routine knowledge-work tasks. Lower cost per task can make repeated document production, financial analysis and business-system workflows more practical, particularly when each job involves multiple model calls or long outputs. The claimed reduction in generated tokens also bears on cost and speed, though the report does not provide latency figures.
The benchmark is aimed at finished work products rather than only short-answer tests. That makes its focus relevant to businesses judging whether a model can prepare a reconciled spreadsheet or a structured presentation, not simply produce plausible prose. At the same time, a score from one company’s internal benchmark cannot establish how well the model performs across all organizations, task types or risk levels. Independent testing and actual operating costs will be needed before buyers can compare the release directly with alternatives for their own workloads.
professional AI document automation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How Paper Instruments Tested It
Paper Instruments presents Ultramarine 1 Flash as a model built for what it calls “professional work,” spanning finance, law, consulting and operations. The company says its benchmark uses professional data and evaluations in model training, and directs readers to a technical report for further detail. No publication date for the report is included in the supplied material, so the precise launch date and development timeline are not established here.
The company’s benchmark chart compares models at different reasoning settings, including low, medium, high and maximum where listed. Those settings matter because both performance and task cost can change with the amount of reasoning used. The chart reports scores and estimated cost per task, while the API price list uses per-million-token rates; the two measures describe different things. Paper Instruments also says its benchmark scores correlate with the Artificial Analysis Intelligence Index at r = 0.81, but that correlation is the company’s characterization of its internal benchmark, not a validation of this model’s score.
The report includes a Nucor example involving five periods of segment sales and earnings. Paper Instruments says the table reconciles its subtotals and totals and shows steel mills earnings falling from $7.2 billion in 2022 to $2.4 billion in 2025, while the segment’s share of sales rose from 58% to 62%. The example illustrates the kind of formatted business analysis the company says the model can produce; it is not, by itself, evidence that the model is reliable across financial reporting tasks.
“An open model for professional work”
— Paper Instruments
AI-powered spreadsheet analysis tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Results Still Needed
The available source is the company’s own product report. It does not include an independent replication, detailed model card, external audit of the benchmark, or enough information to assess how representative the tasks are of everyday business workloads. Although Paper Instruments says each task has a tailored rubric and the benchmark correlates with another index, the supplied material does not show the full task set, scoring examples or uncertainty ranges around the results.
It is also unclear what licensing terms apply to the model, which weights or components are available for use, and whether every capability in the report can be accessed through the API. The report gives token prices and estimated per-task costs but does not spell out the workload assumptions behind each task estimate. Actual expenses may vary with prompt length, reasoning settings, retries and integration requirements. Latency, error rates, data-handling terms and deployment limits are not established in the supplied source.
As an affiliate, we earn on qualifying purchases.
Technical Details and Customer Testing
Paper Instruments points readers to a technical report and a GitHub listing, and presents the Ultramarine API as the route to use the model. The supplied material does not state a release schedule, access restrictions, licensing details or an announced independent evaluation. Those details would help prospective users judge whether the model can be deployed in their own systems and under what terms.
The next useful evidence would include independent benchmark runs, clear documentation of task-cost calculations, and customer tests covering accuracy and reliability on real workflows. Organizations considering the model will also need to check API pricing against their own token use and review security, privacy and integration requirements. Until more evidence is available, the reported score and cost should be treated as Paper Instruments’ claims about its own evaluation, not a guarantee of performance or savings for every user.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Ultramarine 1 Flash?
It is an open model from Paper Instruments designed for professional work in areas such as finance, law, consulting and operations. The company says it is based on GLM-5.3 Flash.
How did it score on the company’s benchmark?
Paper Instruments reports a score of 59.3 with maximum reasoning on its internal PI Knowledge Work benchmark. The supplied material does not include an independent verification of that result.
How much does it cost to use?
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. Its estimated benchmark cost is $0.13 per task, which can differ from a user’s actual bill.
What kinds of tasks does the benchmark cover?
The company says it tests production of Excel workbooks, PowerPoint decks and Word documents, along with work in ERP and manufacturing execution systems. It says rubrics score substance, accuracy and presentation.
Are the performance claims independently confirmed?
Not in the supplied source. The scores, comparison and cost estimates come from Paper Instruments’ report; independent replication and details needed to assess how well the results generalize remain unavailable here.
Source: hn
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
