Paper Instruments

Claude alternatives for professional work

Claude covers writing, analysis, research and code in chat, and agent products carry that work through to finished files. This guide compares nine products on assignments, prices, free plans and review: Claude, ChatGPT, Gemini, Copilot, Perplexity, Mistral’s Vibe, DeepSeek, Manus and Ultramarine Work, which is ours.

Checked on

At a glance

ProductBest forAvailable onStarts atFree access
ClaudeWriting, analysis, research and code in chatWeb, desktop and mobile appsPro $20 a month ($17 billed annually)Yes, limited every 5 hours and weekly
ChatGPTLong multi-step tasks ending in finished filesDesktop app, web and mobilePlus $20 a month (Go $8 limited Work)Yes, limited Work on desktop
Google GeminiErrands, monitoring and research across Gmail and DocsGemini app on web, mobile and MacAI Pro $19.99 a monthYes, chat is free, Task needs Pro or Ultra
Microsoft CopilotAgent work inside Microsoft 365 under IT controlOffice apps, browser, desktop and mobile$21 a user a month plus usage creditsFree consumer app, with limits
Ultramarine WorkProfessional modes with a model trained for the workWeb and desktop app for Mac and Windows$5 a month ($50 a year)7 days free on Starter and Plus
PerplexityCited research and background multi-model runsWeb, Comet, Mac, iOS, Android, SlackPro $20 a month (Max $200)Yes, basic searches and some Computer turns
Vibe (formerly Le Chat)EU-hosted chat, agents and coding on open modelsWeb and mobile, CLI and IDE for codingPro $14.99 a month (Team $24.99)Yes, limited messages and searches
DeepSeekFree reasoning, math and coding on open modelsWeb and mobile apps, APIAPI priced per tokenYes, the app is free
ManusOne-prompt deliverables built in a cloud VMWeb, desktop, mobile and Cue appPro from $20 a monthYes, 300 credits a day

What people use Claude for

Claude is the working chat app for drafting, analysis and code. It runs code and creates files in chat, Claude Sheets (beta) and Claude for Excel add spreadsheets, and Cowork, its file-work agent, is included on paid plans. The free plan covers search, files and code.

Claude alternatives

  1. 1. ChatGPT

    OpenAI

    Good for
    The most widely used assistant at work, with daily writing and analysis in chat. ChatGPT Work, launched July 2026, handles long multi-step tasks that end in a spreadsheet, deck, document or web app, drawing on connected apps such as Gmail, Outlook and GitHub. Codex covers engineering work in the same app.
    How it works
    A chat app with a general agent inside it. Work runs in the desktop app with local files and a browser, plus web and mobile for paid plans.
    Runs on
    OpenAI’s own GPT-5.6 family (Sol, Terra, Luna) at Work’s July 2026 launch, with GPT-6.1 Sol rolling out to Work and Codex from late September 2026.
    Price
    Free and Go ($8) include limited Work, desktop only. Plus is $20 with web and mobile too. Pro runs $100 to $500. Business is $25 a user monthly or $20 annual.
    Choose it when
    You want the broadest ecosystem, with Codex for coding, connected apps such as Gmail, Outlook and GitHub, and a free or $8 route to trying agent work.

    Sources: ChatGPT pricing (OpenAI), ChatGPT Work and Codex (OpenAI help center), ChatGPT plans and usage (OpenAI Learn)

  2. 2. Google Gemini

    Google

    Good for
    The assistant of the Google ecosystem. The Gemini app includes Deep Research with citations, Canvas and Gems free, and Task (formerly Spark) is a personal agent that monitors Gmail, runs errands and creates files across Gmail, Docs, Slides and MCP partners such as Canva, up to 15 at once, on a schedule or trigger.
    How it works
    A personal agent inside the Gemini app, with Chrome auto browse on desktop, a remote browser on mobile, and cloud tasks that run while your laptop is closed.
    Runs on
    Google’s own Gemini 3.5 on the Antigravity agent harness at launch, with Gemini 4 Argon released September 30, 2026, reaching Ultra subscribers first.
    Price
    Task needs a personal Google account, 18 or over, and Google AI Pro at $19.99 or Ultra at $99.99 or $199.99 a month. The free tier has no Task.
    Choose it when
    Your working life runs through Gmail and Docs on a personal account, and you want errands, monitoring and cited research done in the background.

    Sources: Tasks in Gemini (Google help center), Gemini subscription plans (Google)

  3. 3. Microsoft Copilot

    Microsoft

    Good for
    The consumer Copilot app is free with limits. Cowork carries out tasks across Microsoft 365: editing Word, Excel and PowerPoint files in place, sending email, posting in Teams, running scheduled tasks, searching the organization, with approval before important actions, under Purview compliance and admin spend controls.
    How it works
    An agent inside an office suite, working within the Microsoft 365 tenant through Office apps, the browser and desktop or mobile apps, with opt-in role plugins.
    Runs on
    A picker with OpenAI GPT 6.1 Sol and GPT 6 Astra, Claude Opus 5.5, Sonnet 5.5 and Fable 5.1, and Grok 4.7, per Microsoft’s list updated November 5, 2026.
    Price
    The consumer Copilot app is free with limits. Cowork needs a Copilot license (Business $21 a user a month, $18 promo to the end of 2026) plus usage billing.
    Choose it when
    Your firm is standardized on Microsoft 365 and wants agent work inside the tenant, with version history, enterprise governance and IT control over models and spend.

    Sources: Copilot Cowork overview (Microsoft Learn), Copilot Cowork model picker (Microsoft Learn), Microsoft 365 Copilot pricing

  4. 4. Ultramarine Work

    Paper Instruments

    Ultramarine Work is for professionals whose output is files: analysts, accountants, lawyers, consultants. You pick one of over 130 professional modes and stay in it for the workstream, since a mode fixes who the work is for, its recurring deliverables and its skills: Public Markets Research & Investing has twelve skills, from the three-statement method through valuation to the tie-out. It never signs, files, sends or decides, and you review every file it touched.

    Ultramarine Work runs Ultramarine 1 Flash, our model, post-trained on GLM-5.3-Flash for knowledge work. Modes provide the depth: Public Markets Research & Investing serves equity and credit analysts with twelve skills, covering the three-statement method, valuation by comps, precedents and DCF, and a tie-out that traces figures to support. Paper Instruments writes the modes and trains the model, so the two improve together. Ways of working are chosen per task: Standard runs Ultramarine 1 Flash, Assisted adds a frontier model as an advisor, and Advanced runs a frontier model throughout.

  5. 5. Perplexity

    Perplexity

    Good for
    Its center of gravity is research where citations matter. Perplexity Computer breaks an outcome into tasks, runs sub-agents in cloud sandboxes and returns documents, spreadsheets and dashboards. Its own examples price a multi-step research report at $3.50 to $9.50 of credits and a quick fact at $0.15 to $0.70.
    How it works
    A general agent inside a chat app, metered in credits, with an always-on option that runs on your own Mac through the desktop app.
    Runs on
    Third-party models on its own harness: launch-era routing (February 2026) used Claude, Gemini, GPT and Grok per sub-task. Sonar search models are post-trained.
    Price
    Credits cost $1 per 100. Pro ($20) includes a weekly Computer allowance still rolling out and a 4,000-credit bonus. Max ($200) gives 10,000 credits monthly.
    Choose it when
    The citation trail is the deliverable and you want visible per-task accounting.

    Sources: How credits work on Perplexity (help center), Perplexity Computer launch post (Perplexity blog), AI costs and credit examples (Perplexity blog)

  6. 6. Vibe (formerly Le Chat)

    Mistral AI

    Good for
    Vibe is the European all-rounder from Mistral AI, covering chat, a canvas for documents and slides, scheduled tasks and reusable skills, with Vibe Work planning multi-step tasks across connectors and Vibe Code coding in the CLI and IDE. Data is hosted in the EU by default, and you can opt out of training on every plan.
    How it works
    A chat app turned agent for work and code, on web and mobile, with a CLI, VS Code and JetBrains plugins on the coding side.
    Runs on
    Mistral’s own models, several released with open weights. The pricing page does not name the model each plan uses.
    Price
    Free with limited messages, searches and coding sessions. Pro is $14.99 a month (students $5.99) with $25.50 of monthly API credits. Team is $24.99 a user.
    Choose it when
    You need EU hosting or open-weight models, or want a usable free tier.

    Sources: Mistral Vibe product page, Mistral pricing, Where Mistral stores your data (Mistral help center), Opting out of training (Mistral help center)

  7. 7. DeepSeek

    DeepSeek

    Good for
    A free assistant for reasoning, math, coding and general writing, from a lab that trains and releases its own open-weight models. The same models run through a pay-per-token API, which makes DeepSeek a common building block for professionals who assemble their own tools instead of an off-the-shelf app.
    How it works
    A chat app plus a pay-per-token API, with open weights to host yourself.
    Runs on
    DeepSeek’s own open-weight models, with DeepSeek V4.1 Flash featured on its site, served through the API at listed token prices.
    Price
    The app is free and there is no consumer subscription. The API is pay-per-token (Flash input $0.15 per million tokens), and the open weights cost only hosting.
    Choose it when
    You would rather run the model yourself, wire it into your own tools, or compare answers at no cost.

    Sources: DeepSeek official site, DeepSeek API pricing

  8. 8. Manus

    Manus

    Good for
    Manus turns one prompt into a hands-off deliverable: it plans and executes in a cloud virtual machine, returning sites, slide decks, research reports and video, with event-triggered automations, an always-on Cloud Computer and up to 20 concurrent tasks on paid plans. Credits are refunded when a task fails on its side.
    How it works
    A general agent with its own cloud computer and in-house harness (Cascade), controlled from web, desktop, mobile and the Cue personal-agent app.
    Runs on
    Own Cascade agent harness. The company does not name its current core model, and its models page says it picks a media model for each task, or lets you choose.
    Price
    Free with 300 credits a day and one concurrent task. Pro starts at $20 a month for 4,000 credits, an 8,000-credit tier from $40, Team from $20 a seat.
    Choose it when
    The job is self-contained and broad, such as building a site or running parallel research to a finished file, and you would rather review the output at the end.

    Sources: Manus 2.0 announcement (Manus blog), Manus membership pricing (help center)

Pricing and how to try it

Plans start at $5 a month, and Starter and Plus begin with 7 days free. On our internal PI Knowledge Work benchmark, Ultramarine 1 Flash costs about $0.13 a task, about a tenth of Claude Opus 5.5 or GPT-6 Sol at comparable scores.

Start free trial See every plan

Questions

Is there a free way to try Ultramarine Work?

Starter and Plus begin with 7 days free, then cost $5 and $20 a month, and there is no free tier. Most of the other products here offer ongoing free access: Claude’s free plan includes search, file creation and code execution within limits, ChatGPT runs limited Work on desktop for free accounts, DeepSeek’s app is free, and Manus, Vibe, Gemini and Perplexity all have free tiers, with the agentic parts usually the most restricted.

Should I subscribe to Claude instead?

If you want Claude itself, Anthropic’s plans are the direct place to get it, and that is the honest starting point. What Ultramarine Work adds is the professional layer around the model: a mode stays selected for the workstream and configures the agent with curated skills such as the three-statement method, the valuation math and the tie-out review, your firm’s style and over 50 connectors apply to every task, and the finished file arrives as a draft you review.

How is a mode different from a plugin or a custom bot?

A mode configures a whole practice for a workstream: who the work is for, its recurring deliverables, and a set of domain skills with their methods, review standards and common traps. A plugin is an extension you enable in a general agent, and a well-built one bundles real skills that load when a task needs them. The difference is scope and persistence: you select a mode once and it shapes every step of the workstream, while a plugin adds its skills to a general agent, and plugins vary in depth.

Are the numbers in the Taloncrest example real?

They are not, and they are not meant to be investment data. Taloncrest Industries, its filings, its transcript and every figure in the demo are invented for illustration, with the arithmetic made to add up so you can see the shape of the work. The product facts describe features available on the checked date of 2026-11-10.

Who checks the work before it goes out?

The person using the tool does. The agent builds drafts, models and files, and its skills have it trace figures to support and log exceptions before it hands everything over, with every file it touched left for review. Ultramarine Work never signs, files, sends, posts to a system of record or makes the professional decision, and the tie-out step exists so the review starts from a list of known exceptions rather than a polished document.

What happens to my data when I use it?

Ultramarine Work changes files only in the project’s folder. When a request needs it, it can read files elsewhere on your computer, never passwords, keys or app data, and you can turn that off in settings. The model runs on Paper Instruments’ servers, so the content it works on is sent there to be processed. Paper Instruments is SOC 2 Type II, and our security and compliance details are on our compliance page.

Sources

Checked on