(ultramarine/logos/pi.svg)  (ultramarine/logos/gemini.svg)  (ultramarine/logos/deepseek.svg)  (ultramarine/logos/claude.svg)  (ultramarine/logos/zai.svg)  (ultramarine/logos/kimi.svg)  (ultramarine/logos/openai.svg)  (feather/powerpoint.svg)  (_next/static/css/2f46e0db0d4a3aed.css@dpl=dpl_4EtRUbHFCsQGWq1Cmx29hJ5WdTqc.css)  (_next/static/css/43344b939570a3c1.css@dpl=dpl_4EtRUbHFCsQGWq1Cmx29hJ5WdTqc.css)  (_next/static/css/a26735b9e621f7be.css@dpl=dpl_4EtRUbHFCsQGWq1Cmx29hJ5WdTqc.css)  (_next/static/css/429664b0dc6fb195.css@dpl=dpl_4EtRUbHFCsQGWq1Cmx29hJ5WdTqc.css)  (_next/static/css/56351ed819551fb1.css@dpl=dpl_4EtRUbHFCsQGWq1Cmx29hJ5WdTqc.css)  (_next/static/css/de28fea26c9020dc.css@dpl=dpl_4EtRUbHFCsQGWq1Cmx29hJ5WdTqc.css)  (https://www.paperinstruments.com/_next/static/chunks/webpack-8b50438a1751653e.js?dpl=dpl_4EtRUbHFCsQGWq1Cmx29hJ5WdTqc)  Ultramarine 1 Flash — Paper Instruments   (https://www.paperinstruments.com)   (manifest.webmanifest)         (ultramarine.html)                (favicon.svg)  (favicon.svg)    (https://www.paperinstruments.com/)                       Paper Instruments  (ultramarine.html) Ultramarine (https://www.paperinstruments.com/feather) Feather (https://www.paperinstruments.com/paper-python) Paper Office (https://www.paperinstruments.com/blog) Research     Ultramarine 1Flash An open model for professional work Ultramarine 1 Flash is Fable-class on knowledge work spanning finance, law, consulting, and operations. It is also 60x cheaper than Fable. (https://platform.paperinstruments.com) Ultramarine API →  (https://github.com/paper-instruments/feather-train)       Model weights coming soon    (https://www.paperinstruments.com/blog/ultramarine) Technical report →    Built for professional work JobBench ()  Ultramarine 1 Flash   62.9  ()  Gemini 3.8 Flash   60.6  ()  DeepSeek V4.1 Flash   60.4  ()  Claude Fable 5   57.4  ()  GLM-5.3 Flash   56.3  ()  Kimi K3   54.3  ()  GPT-5.6 Sol   45.4    PI Knowledge Work (internal) ()  Ultramarine 1 Flash   59.3  ()  Claude Fable 5   57.8  ()  GPT-5.6 Sol   55.1  ()  GLM-5.3 Flash   52.0  ()  Kimi K3   51.2  ()  Gemini 3.8 Flash   49.4  ()  DeepSeek V4.1 Flash   47.9    FrontierFinance ()  Ultramarine 1 Flash   53.3  ()  GLM-5.3 Flash   50.7  ()  Gemini 3.8 Flash   50.0  ()  Claude Fable 5   49.2  ()  GPT-5.6 Sol   46.8  ()  Kimi K3   46.4  ()  DeepSeek V4.1 Flash   46.0    AutomationBench ()  DeepSeek V4.1 Flash   51.3  ()  Ultramarine 1 Flash   50.2  ()  GLM-5.3 Flash   48.8  ()  Gemini 3.8 Flash   47.3  ()  Kimi K3   46.7  ()  Claude Fable 5   46.2  ()  GPT-5.6 Sol   45.8    DiligenceBench ()  Gemini 3.8 Flash   53.1  ()  Claude Fable 5   52.9  ()  Ultramarine 1 Flash   50.6  ()  Kimi K3   50.4  ()  DeepSeek V4.1 Flash   50.3  ()  GPT-5.6 Sol   50.0  ()  GLM-5.3 Flash   48.2    APEX-Agents 1.1 ()  Gemini 3.8 Flash   64.3  ()  Claude Fable 5   63.6  ()  Ultramarine 1 Flash   57.0  ()  GLM-5.3 Flash   52.8  ()  GPT-5.6 Sol   51.5  ()  Kimi K3   50.6  ()  DeepSeek V4.1 Flash   39.5       Improved token efficiency Across the six benchmarks, Ultramarine 1 Flash generates 6.5% fewer tokens than its base, GLM-5.3 Flash, taking a shorter path to better work. ()  GLM-5.3 Flash 32.7M    ()  Ultramarine 1 Flash 30.5M     −6.5%   Output tokens generated across all six benchmarks, reasoning included. Where GLM-5.3 Flash’s score is a public one, its tokens come from our own runs closest to that score.   Fable-class model for cents per task Ultramarine 1 Flash sits on the cost-quality frontier at the most attractive price point. At $0.13 per task, it outscores Claude Fable 5 (max) and GPT-5.6 Sol (max), at $8.19 and $2.09 per task, and comes within three points of Claude Opus 5.5 (max) and GPT-6 Sol (max) at $6.96 and $1.26 per task. We measure this on PI Knowledge Work, a carefully built internal benchmark whose scores correlate strongly with the Artificial Analysis Intelligence Index (r = 0.81). Input $0.15  Cached input $0.03  Output $0.50   Per million tokens on the Ultramarine API, the same as GLM-5.3 Flash.     42   46   50   54   58   62   66   $0.01   $0.03   $0.1   $0.3   $1   $3   $10   Cost per task (log scale) Score  GLM-5.3 Flash, max reasoning: 52.0 at $0.10 per task   GPT-6 Astra, low reasoning: 60.3 at $1.22 per task   GPT-6 Astra, medium reasoning: 62.4 at $1.90 per task   GPT-6 Astra, max reasoning: 64.2 at $5.64 per task   Claude Opus 5.5, low reasoning: 53.9 at $0.49 per task   Claude Opus 5.5, medium reasoning: 57.0 at $1.10 per task   Claude Opus 5.5, high reasoning: 61.4 at $1.53 per task   Claude Opus 5.5, max reasoning: 61.9 at $6.96 per task   GPT-6 Sol, low reasoning: 55.3 at $0.15 per task   GPT-6 Sol, medium reasoning: 55.5 at $0.26 per task   GPT-6 Sol, high reasoning: 56.5 at $0.42 per task   GPT-6 Sol, max reasoning: 61.4 at $1.26 per task   Claude Fable 5, low reasoning: 50.8 at $1.82 per task   Claude Fable 5, medium reasoning: 48.6 at $2.85 per task   Claude Fable 5, max reasoning: 57.8 at $8.19 per task   Claude Fable 5.1, max reasoning: 65.5 at $8.64 per task   GPT-5.6 Sol, low reasoning: 48.8 at $0.42 per task   GPT-5.6 Sol, medium reasoning: 53.4 at $0.88 per task   GPT-5.6 Sol, high reasoning: 50.5 at $1.28 per task   GPT-5.6 Sol, max reasoning: 55.1 at $2.09 per task   Muse Spark 1.3, low reasoning: 51.2 at $0.48 per task   Muse Spark 1.3, medium reasoning: 50.7 at $0.98 per task   Muse Spark 1.3, high reasoning: 55.3 at $1.09 per task   Muse Spark 1.3, max reasoning: 54.7 at $1.58 per task   GPT-6 Luna, low reasoning: 43.2 at $0.008 per task   GPT-6 Luna, medium reasoning: 54.1 at $0.035 per task   GPT-6 Luna, high reasoning: 53.1 at $0.054 per task   GPT-6 Luna, max reasoning: 58.3 at $0.11 per task   Gemini 3.8 Flash, low reasoning: 45.6 at $0.35 per task   Gemini 3.8 Flash, medium reasoning: 51.0 at $0.96 per task   Gemini 3.8 Flash, high reasoning: 49.4 at $1.18 per task   DeepSeek V4.1 Flash, max reasoning: 47.9 at $0.35 per task   Kimi K3, max reasoning: 51.2 at $1.33 per task   GLM-5.3, max reasoning: 54.1 at $1.27 per task   DeepSeek V4 Pro, max reasoning: 54.1 at $0.43 per task   Ling 3.1 Flash, reasoning: 51.6 at $0.21 per task   Qwen3.8 2.4T, xhigh reasoning: 50.6 at $0.94 per task   Qwen3.8 27B, xhigh reasoning: 50.1 at $0.42 per task   MiniMax M3, reasoning: 44.2 at $0.24 per task    Ultramarine 1 Flash, max reasoning: 59.3 at $0.13 per task   Ultramarine 1 Flash(max)   GLM-5.3 Flash(max)   GPT-6 Astra(low)   GPT-6 Astra(medium)   GPT-6 Astra(max)   Claude Opus 5.5(low)   Claude Opus 5.5(medium)   Claude Opus 5.5(high)   Claude Opus 5.5(max)    GPT-6 Sol(low)   GPT-6 Sol(medium)   GPT-6 Sol(high)   GPT-6 Sol(max)   Claude Fable 5(low)   Claude Fable 5(medium)   Claude Fable 5(max)    Claude Fable 5.1(max)   GPT-5.6 Sol(low)   GPT-5.6 Sol(medium)   GPT-5.6 Sol(high)   GPT-5.6 Sol(max)   Muse Spark 1.3(low)    Muse Spark 1.3(medium)   Muse Spark 1.3(high)   Muse Spark 1.3(max)   GPT-6 Luna(low)   GPT-6 Luna(medium)   GPT-6 Luna(high)   GPT-6 Luna(max)   Gemini 3.8 Flash(low)    Gemini 3.8 Flash(medium)   Gemini 3.8 Flash(high)   DeepSeek V4.1 Flash(max)   Kimi K3(max)   GLM-5.3(max)   DeepSeek V4 Pro(max)    Ling 3.1 Flash(reasoning)   Qwen3.8 2.4T(xhigh)   Qwen3.8 27B(xhigh)   MiniMax M3(reasoning)     PI Knowledge Work is our internal benchmark of professional work in finance, law, consulting, and operations. Its tasks ask the model to deliver finished Excel workbooks, PowerPoint decks, and Word documents, and to work inside live business systems such as ERP and MES. A rubric written for each task scores the result on substance, accuracy, and presentation. Every model works in the same harness, and costs are at each developer's official API pricing. Read our(https://www.paperinstruments.com/blog/ultramarine) technical report for more on the kind of professional data and evaluations that trained this model.   Crafted with a sense of client readiness Ultramarine 1 Flash writes clean, simple, straightforward prose and has a strong sense of visual and information hierarchy. It is built to adhere to the same rules required of analysts in professional services. (Slide with a table of Nucor segment sales and earnings from 2022 to the first half of 2026, a sales mix bar, and notes on what moves the result)  ()  ()  ()  ()  ()   (PowerPoint)  Nucor through the steel cycle Five periods of segment sales and earnings in one table, with every subtotal and total reconciling. Steel mills earnings fell from $7.2bn in 2022 to $2.4bn in 2025, while their share of sales rose from 58% to 62%.  ()    ()    ()    ()    ()    ()        (https://platform.paperinstruments.com) Ultramarine API →  (https://www.paperinstruments.com/feather) Feather →  (https://www.paperinstruments.com/blog/ultramarine) Technical report →  (https://github.com/paper-instruments/feather-train) GitHub ↗  Hugging Face Model weights coming soon    © 2026Paper Instruments, Inc. (https://docs.paperinstruments.com/) Docs (https://www.paperinstruments.com/careers) Careers (https://www.paperinstruments.com/careers) Contact (https://github.com/paper-instruments) GitHub  (https://www.paperinstruments.com/privacy) Privacy (https://www.paperinstruments.com/terms) Terms (https://compliance.paperinstruments.com/) SOC 2 Type II