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 professional knowledge work. The company reports a 59.3 score on its internal PI Knowledge Work benchmark at an estimated $0.13 per task, while noting the results come from its own evaluation. Independent validation and further details about the model’s release remain unclear.
Paper Instruments has introduced Ultramarine 1 Flash, an open model designed for finance, law, consulting and operations, and says it scored 59.3 on its internal PI Knowledge Work benchmark at an estimated cost of $0.13 per task. The release matters to organizations weighing the cost of AI systems for document and business-process work, but the headline performance figures are the company’s own benchmark results, not independent confirmation.
The company says Ultramarine 1 Flash is based on GLM-5.3 Flash and generated 6.5% fewer tokens than that base model across six benchmarks, including reasoning tokens. Paper Instruments says the comparison used its own model runs for token counts, while the GLM-5.3 Flash score shown in its chart was a public score. The supplied material does not identify the six benchmarks individually.
On PI Knowledge Work, Paper Instruments reports a score of 59.3 in maximum-reasoning mode, at $0.13 per task. Its chart lists Claude Fable 5 at 57.8 and $8.19 per task, GPT-5.6 Sol at 55.1 and $2.09, Claude Opus 5.5 at 61.9 and $6.96, and GPT-6 Sol at 61.4 and $1.26. These are the company’s benchmark and task-cost estimates, not a guarantee of equivalent performance or spend in other workloads.
Paper Instruments lists Ultramarine API prices of $0.15 per million input tokens, $0.03 per million cached input tokens and $0.50 per million output tokens, matching the rates it gives for GLM-5.3 Flash. The company describes the model as open, but the material provided does not specify the license or the precise terms for accessing or modifying its weights.
The Cost of Business AI
The announcement is relevant to teams using AI for tasks that require finished documents or work in business software, where both output quality and the cost of repeated model calls can affect adoption. Paper Instruments says its benchmark covers finance, law, consulting and operations, and measures work such as creating Excel workbooks, PowerPoint presentations and Word documents, as well as acting inside ERP and manufacturing execution systems.
If the reported results hold up in independent testing and practical deployments, the low listed task cost could make more capable models accessible for routine professional workflows. But the benchmark score alone does not establish how well the model handles a particular company’s data, instructions, security requirements or error tolerance. Readers should treat the price and quality comparison as a company-reported starting point, not a universal cost-benefit result.
AI language model API for business
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How Paper Instruments Tested It
Paper Instruments describes PI Knowledge Work as an internal benchmark built around professional tasks. It says each task has a rubric covering substance, accuracy and presentation, that models run through the same harness, and that costs use each developer’s official API pricing. The company says the benchmark’s scores correlate strongly with the Artificial Analysis Intelligence Index, reporting a correlation coefficient of r = 0.81.
That correlation is a claim about the relationship between the internal benchmark and another index; it does not, by itself, independently verify Ultramarine’s score. The release also says the model was trained using professional data and evaluations, but the supplied material does not give detailed information about the training corpus. Paper Instruments characterizes its writing as straightforward and its outputs as attentive to visual and information hierarchy, and shows an example involving Nucor’s segment sales and earnings.
“An open model for professional work”
— Paper Instruments
professional AI document generation tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Results and Access Terms
The supplied material does not include an independent evaluation of Ultramarine 1 Flash or enough detail to reproduce the company’s benchmark results. It is also unclear how PI Knowledge Work tasks were selected, how many were tested, how scores vary across individual tasks, or whether the reported $0.13 average reflects particular settings and usage assumptions. The company says its score correlates with the Artificial Analysis Intelligence Index, but does not provide the underlying analysis here.
Other unresolved points include the model’s exact open-source or open-weight license, where its weights can be obtained, availability limits, and what privacy and data-handling terms apply to API users. The announcement does not establish how the model performs against confidential or specialized business data, or whether its outputs meet legal, financial or operational requirements without human review.
As an affiliate, we earn on qualifying purchases.
Validation and Release Details
Paper Instruments points readers to a technical report and GitHub, but the supplied information does not state a publication schedule, a specific release date or a next benchmark milestone. Further details about access, licensing and model deployment would help developers judge whether they can test it in their own systems.
For now, organizations considering the model can compare the published API rates with their expected workloads and seek task-level tests before relying on the benchmark’s overall score. Independent evaluations, clearer reporting on the benchmark methodology and documentation of the model’s release terms would help establish how closely the company’s claims translate to real professional use.
cost-effective AI for legal and finance tasks
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Ultramarine 1 Flash?
Paper Instruments describes it as an open model for professional work, aimed at areas including finance, law, consulting and operations.
What score did Paper Instruments report?
The company reports a score of 59.3 in maximum-reasoning mode on its internal PI Knowledge Work benchmark. The result has not been independently confirmed in the supplied material.
How much does it cost to use?
Paper Instruments estimates $0.13 per benchmark task. Its API price list gives $0.15 per million input tokens, $0.03 per million cached input tokens and $0.50 per million output tokens.
Is the model open source?
The company calls Ultramarine 1 Flash an open model, but the supplied announcement does not specify its license or explain exactly what model files and modification rights are available.
Has an independent evaluator verified the results?
No independent validation is identified in the supplied material. The reported score, cost comparisons and benchmark description are attributed to Paper Instruments.
Source: hn
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
