--- title: Ultramarine 1 Flash: an open model for finance, law and... description: Paper Instruments released Ultramarine 1 Flash, an open-source model post-trained from GLM-5.3 Flash for finance, law and consulting document work such as... canonical: https://daily.dev/posts/ultramarine-1-flash-an-open-model-for-finance-law-and-consulting-documents-1wzwyfj4g twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Ultramarine 1 Flash: an open model for finance, law and consulting documents | daily.dev og:description: Paper Instruments released Ultramarine 1 Flash, an open-source model post-trained from GLM-5.3 Flash for finance, law and consulting document work such as... og:url: https://daily.dev/posts/ultramarine-1-flash-an-open-model-for-finance-law-and-consulting-documents-1wzwyfj4g og:image: https://api.daily.dev/og/posts/1WZwYFJ4g.png og:image:alt: Ultramarine 1 Flash: an open model for finance, law and consulting documents og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Ultramarine 1 Flash: an open model for finance, law and consulting documents **[Collections](https://daily.dev/sources/collections)** · 2 min read · 2 upvotes · 0 comments ## Summary Paper Instruments released Ultramarine 1 Flash, an open-source model post-trained from GLM-5.3 Flash for finance, law and consulting document work such as Excel models, PowerPoint decks and Word memos. On the company's own PI Knowledge Work benchmark it scored 59.3 at $0.13 per task, edging out Claude Fable 5 (max) at 57.8 points but $8.19 per task, and landing within three points of Claude Opus 5.5 (max) and GPT-6 Sol (max) at a fraction of the cost. API pricing mirrors the base model at $0.15/M input tokens, $0.03/M cached input tokens and $0.50/M output tokens. ## Content Paper Instruments has released Ultramarine 1 Flash, an open model aimed at document-heavy work at banks, law firms and consultancies. It produces finished Excel models, PowerPoint decks and Word memos, and the company says each task costs cents. ## What it is Ultramarine 1 Flash is a post-training of GLM-5.3 Flash. Paper Instruments describes it as frontier-class for knowledge work in finance, law, consulting and operations. It generates 6.5% fewer tokens than the base model. Training tasks were each graded against their own rubric. A diligence memo or a board deck has no single correct answer, so a fixed answer key wouldn't work. ## Benchmark results The numbers come from Paper Instruments' own PI Knowledge Work benchmark, which measures client-ready deliverables. That makes them the company's claims, not an independent test. - Ultramarine 1 Flash scored 59.3 at $0.13 per task. - Claude Fable 5 (max) scored 57.8 at $8.19 per task. - It also scores higher than GPT-5.6 Sol (max). - It lands within three points of Claude Opus 5.5 (max) and GPT-6 Sol (max), at a fraction of their per-task cost. The cost gap is the interesting part. A benchmark lead over Claude Fable 5 of 1.5 points is small, but doing it for $0.13 instead of $8.19 is a very different proposition for anyone running thousands of documents. ## Pricing API pricing matches the base model: - $0.15 per million input tokens - $0.03 per million cached input tokens - $0.50 per million output tokens The model is open-source. ## Questions this post answers ### How does Ultramarine 1 Flash compare to Claude Fable 5 on cost and performance for document generation? Ultramarine 1 Flash scored 59.3 on Paper Instruments' PI Knowledge Work benchmark at $0.13 per task, versus 57.8 for Claude Fable 5 (max) at $8.19 per task. It also lands within three points of Claude Opus 5.5 (max) and GPT-6 Sol (max) while costing a fraction as much, making it notably cheaper for high-volume document generation like financial models and legal memos. _Teams weighing cost against quality for document-generation pipelines can track model comparisons like this on daily.dev._ ### What base model is Ultramarine 1 Flash built on and how is it priced? Ultramarine 1 Flash is a post-training of GLM-5.3 Flash, generating 6.5% fewer tokens than the base model. API pricing matches the base model exactly: $0.15 per million input tokens, $0.03 per million cached input tokens, and $0.50 per million output tokens. It is released as an open-source model. _Developers evaluating open-source LLM options for production use can follow pricing and release details on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 2 discussions and 101 comments across hackernews (as of 2026-10-17). **TL;DR:** The dominant thread of discussion criticizes the benchmark/marketing presentation (cherry-picked comparisons, misleading cost-per-task charts) while a smaller contingent appreciates the price/performance tradeoff and discusses tooling/harnesses for using the model in practice. **Sentiment:** 20% positive · 40% mixed · 40% skeptical **The case for** - Several commenters find the price-to-performance ratio genuinely compelling given it's dramatically cheaper than frontier models. - Some note that applying large-scale reinforcement learning to an open base model is non-trivial and more substantive than typical 'skinned' fine-tunes. - A few are enthusiastic about distillation-style models putting competitive pressure on larger incumbent labs. **The pushback** - Many argue the benchmark graphs are misleadingly cherry-picked, comparing only against weaker models in one chart and conveniently reintroducing stronger/pricier models only in the cost chart. - Some say they'd dismiss the model outright because of the perceived statistical tricks in its marketing. - Others question whether 'frontier' claims are overstated, noting its performance looks closer to a mid-tier model than to true frontier models. - Concerns raised about practical switching costs and risk for non-technical users compared to staying with an established provider. - Some doubt the real-world usefulness of benchmarks in general, including this one. **By community** - hackernews (skeptical): Discussion leans critical, centering heavily on accusations of misleading benchmark presentation, though some defend the cost/performance value and discuss practical tooling and harness options. - hackernews (mixed): A minimal secondary thread just links to a launch video with no substantive discussion. **Hottest debate:** Whether the benchmark and cost-per-task charts are legitimate or deliberately cherry-picked to flatter the model's positioning. **Open questions** - How exactly is the model able to be dramatically cheaper while reaching close-to-frontier scores — is the cost comparison apples-to-apples? - What harness or tooling is actually needed to get the model producing real Excel/PowerPoint/Word deliverables in practice? - Would non-technical teams realistically switch from established providers given setup and ongoing-maintenance risk? **Highlights** > As an aside, I do not like unnecessarily misleading graphs. Own your strengths and weaknesses. In the "Results" section, they choose to compare Ultramarine only to models that it's better than on every metric. They ignore models like Fable 5.1, GPT-6 Astra, Claude Opus 5.5, etc. Then, in the "Score against cost per task" graph, which is focused on price, suddenly all of those models that perform better than Ultramarine come roaring back because they are also more expensive. Just list yourselves as the second best sometimes. Your thing is impressive. Don't let the idiot marketing guy hide that behind stupid tricks. > — [CobrastanJorji on hackernews · 8 comments](https://news.ycombinator.com/item?id=50128230) > Yes, for me when they do this I immediately dismiss the model, don't even bother testing it. If you need to lie with statistics your model isn't good enough. > — [digitcatphd on hackernews](https://news.ycombinator.com/item?id=50129046) > Lol this is not Fable-class. This is Luna class. Look at the graphs. > — [Kevcmk on hackernews](https://news.ycombinator.com/item?id=50128316) > I’m confused this seems exactly like a skinned glm 5.3 - flash with a few context injection tricks and guardrails - whats going on with these “startups”? > — [bythreads on hackernews · 3 comments](https://news.ycombinator.com/item?id=50128568) > As someone who is not a software developer (there are some of us here, lurking around because we learn things) but who runs a business and spends his days doing said 'knowledge work', how and why would something like this convince me to move the company away from using Claude (as a desktop app)? I'm sure these models are cheaper to use than our Team subscription with Claude, but it feels like it would be a massive headache for me to work out how to select and implement some harness for this and set it up for people, and the space is moving so quickly I'd feel I need to keep considering when to move to some new model, introducing risk. The safest bet for us and most companies I expect is to stick with what you know works and pay a premium for that. Thoughts very welcome please. > — [binbag on hackernews · 1 comments](https://news.ycombinator.com/item?id=50129029) **Source threads** - [hackernews](https://news.ycombinator.com/item?id=50106747) · 260 points · 99 comments - [hackernews](https://news.ycombinator.com/item?id=50092655) · 24 points · 2 comments ## Similar posts on daily.dev - [Google's Gemini 3.5 Flash costs 3x the model it replaced, and the era of cheap AI is ending](https://daily.dev/posts/google-s-gemini-3-5-flash-costs-3x-the-model-it-replaced-and-the-era-of-cheap-ai-is-ending-ochkeykzb) · XDA Developers · 2 upvotes · 0 comments --- Tags: [#open-source](https://daily.dev/tags/open-source), [#llm](https://daily.dev/tags/llm), [#fintech](https://daily.dev/tags/fintech), [#glm](https://daily.dev/tags/glm) [View this post on daily.dev](https://daily.dev/posts/ultramarine-1-flash-an-open-model-for-finance-law-and-consulting-documents-1wzwyfj4g) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Ultramarine 1 Flash: an open model for finance, law and consulting documents","url":"https://daily.dev/posts/ultramarine-1-flash-an-open-model-for-finance-law-and-consulting-documents-1wzwyfj4g","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/ultramarine-1-flash-an-open-model-for-finance-law-and-consulting-documents-1wzwyfj4g"},"datePublished":"2026-10-17T05:11:02.353Z","dateModified":"2026-10-17T11:41:27.032Z","description":"Paper Instruments released Ultramarine 1 Flash, an open-source model post-trained from GLM-5.3 Flash for finance, law and consulting document work such as...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/4b593c4a6be3ef4418237afbc67f8d6f?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/4b593c4a6be3ef4418237afbc67f8d6f?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/ultramarine-1-flash-an-open-model-for-finance-law-and-consulting-documents-1wzwyfj4g","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"open-source,llm,fintech,glm","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"Ultramarine 1 Flash: an open model for finance, law and consulting documents"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/ultramarine-1-flash-an-open-model-for-finance-law-and-consulting-documents-1wzwyfj4g#faq","mainEntity":[{"@type":"Question","name":"How does Ultramarine 1 Flash compare to Claude Fable 5 on cost and performance for document generation?","acceptedAnswer":{"@type":"Answer","text":"Ultramarine 1 Flash scored 59.3 on Paper Instruments' PI Knowledge Work benchmark at $0.13 per task, versus 57.8 for Claude Fable 5 (max) at $8.19 per task. It also lands within three points of Claude Opus 5.5 (max) and GPT-6 Sol (max) while costing a fraction as much, making it notably cheaper for high-volume document generation like financial models and legal memos. Teams weighing cost against quality for document-generation pipelines can track model comparisons like this on daily.dev."}},{"@type":"Question","name":"What base model is Ultramarine 1 Flash built on and how is it priced?","acceptedAnswer":{"@type":"Answer","text":"Ultramarine 1 Flash is a post-training of GLM-5.3 Flash, generating 6.5% fewer tokens than the base model. API pricing matches the base model exactly: $0.15 per million input tokens, $0.03 per million cached input tokens, and $0.50 per million output tokens. It is released as an open-source model. Developers evaluating open-source LLM options for production use can follow pricing and release details on daily.dev."}}]} ```