Paper Instruments

Claude Cowork alternatives for professional work

Claude Cowork delegates multi-step work across files, apps and connectors, and teams extend it with finance plugins whose investment banking skills cover CIMs and merger models. Seven alternatives follow, and each entry states what the product reaches, its models, price and how a finished file gets reviewed.

Checked on

At a glance

ProductBest forAvailable onStarts atFree access
Claude CoworkMulti-step work in your files, apps and connectorsDesktop (macOS, Windows), web and mobile (rolling out)$20 a month (Pro)None, paid plans only
ChatGPT WorkLong tasks ending in files, built from connected appsDesktop, web and mobile (desktop only on Free and Go)$20 a month (Plus)Limited desktop access on Free and Go
Microsoft 365 Copilot CoworkAgent work inside Microsoft 365 under IT governanceBrowser, Copilot desktop and mobile, Microsoft 365 apps$21 a user a month (Business) plus usageNone, needs a Copilot license
Perplexity ComputerResearch jobs that end in cited filesWeb, mobile, Comet browser, Mac app$20 a month (Pro)Free search, limited Computer turns
Ultramarine WorkProfessional deliverables built in a mode for the practiceWeb and desktop app for Mac and Windows$5 a month (Starter, 3,000 credits)7 days free on Starter and Plus
ManusHands-off deliverables from one promptWeb, desktop and mobile$20 a month (Pro, 4,000 credits)Free plan, 300 credits a day
Gemini TaskPersonal tasks across Gmail, Docs and partner appsGemini web, mobile and Mac apps$19.99 a month (Google AI Pro)None, needs Pro or Ultra, personal account
GensparkQuick slides, sheets and sites from one promptWeb, mobile, Chrome sidebar$24.99 a month (Plus, 10,000 credits)About 100 credits a day, capped lifetime
Grok BotRoutine ops run by named bots on cloud computersDesktop, mobile, X, Slack$30 a month (The X Subscription Premium)None, Premium and above or legacy Cursor

What people use Claude Cowork for

You hand Cowork a goal and it works across your files, apps and connectors: organizing a folder, producing reports and decks, and running scheduled jobs while the laptop is closed. Anthropic’s finance plugins, one for investment banking, add CIM, teaser and merger-model skills.

Claude Cowork alternatives

  1. 1. ChatGPT Work

    OpenAI

    Good for
    Long, multi-step jobs that end in a finished file: pulling data from Salesforce or Gmail, checking Slack for context, then writing the deck, sheet or small web app. It reaches the web and connected apps, pauses for approval before consequential actions, and supports a large plugin catalog.
    How it works
    A general agent inside the ChatGPT app, in a rebuilt desktop app that hosts Chat, Work and Codex, with web and mobile on paid plans.
    Runs on
    OpenAI’s own GPT-5.6 Sol, Terra and Luna at launch (2026-07-09), with GPT-6.1 Sol rolling out in Work from 2026-09-29, starting with Pro plans.
    Price
    Included in ChatGPT plans. Free and Go ($8) get limited desktop-only access, Plus is $20 a month, Pro starts at $100, Business is $25 a seat monthly.
    Choose it when
    You already pay for ChatGPT and want long sessions across the web and connected apps, or want to test agent work on the free desktop access first.

    Sources: ChatGPT pricing (Learn), ChatGPT Work and Codex help

  2. 2. Microsoft 365 Copilot Cowork

    Microsoft

    Good for
    Delegated work inside the Microsoft 365 tenant: drafting email, scheduling meetings, editing shared Word, Excel and PowerPoint files in place with version history, posting to Teams, and scheduled jobs, under Purview compliance, audit logs and admin spending controls, with approval before each important action.
    How it works
    An agent inside the office suite: browser, Copilot desktop and mobile apps, and the Microsoft 365 apps themselves, under tenant-wide admin policies.
    Runs on
    Uses Anthropic’s Cowork harness. Model picker with Auto, GPT 6.1 Sol, GPT 5.6 Terra, Claude Opus 5.5, Fable 5.1 and Grok 4.7 (November 2026).
    Price
    A Copilot license plus usage-based billing: Enterprise $30 a user monthly, Business $21 ($18 promo to 2026-12-31 for existing eligible customers).
    Choose it when
    Your organization runs on Outlook, Teams, SharePoint and shared Office files, and IT needs edits in place with version history, Purview compliance and spending controls.

    Sources: Copilot Cowork models (Microsoft Learn), Copilot Cowork overview (Microsoft Learn)

  3. 3. Perplexity Computer

    Perplexity

    Good for
    Research-heavy jobs where a source must sit behind every claim: market sizing, competitive screens and diligence memos that end in cited documents, spreadsheets, dashboards or apps. Tasks run as sub-agents in cloud sandboxes, continue in the background or on a schedule, and report the credits each thread used.
    How it works
    A general digital worker that breaks a goal into tasks and runs sub-agents in cloud sandboxes, plus a Mac app, Personal Computer, that works on local files.
    Runs on
    Third-party models on Perplexity’s own harness: Claude Opus 4.6 for core reasoning at launch (2026-02-25), reportedly GPT-5.6 Terra and Luna defaults since.
    Price
    Pro $20 a month with weekly Computer usage still rolling out (without credits the Light tier only), Max $200 with 10,000 credits monthly, credits at 100 for $1.
    Choose it when
    Someone who did not run the task will check it, so every claim needs a citation trail, or you want to compare outputs across several labs’ models.

    Sources: How credits work on Perplexity, Perplexity Computer launch post

  4. 4. Ultramarine Work

    Paper Instruments

    Ultramarine Work is an AI coworker with the profession built in. You pick a practice, such as Investment Banking, and the agent works with that practice’s methods and review standards for the whole workstream: 14 skills from comps and DCFs to merger models, process materials and the tie-out. Bankers and other deal professionals are its closest fit. Teams that live in Microsoft 365 fit Copilot Cowork, and hands-off cloud jobs fit Manus or Perplexity Computer.

    Ultramarine Work pairs a written professional library with a model trained for the same work. Ultramarine 1 Flash is post-trained on GLM-5.3-Flash. The Investment Banking mode loads 14 skills written as methods, with reviewer standards and common traps. The agent picks what a step needs, cites the file behind each figure and notes what it could not reconcile. Standard runs Ultramarine 1 Flash, Assisted adds a frontier model as the task’s advisor, and Advanced hands a frontier model the whole task. Paper Instruments writes the skills and trains the model, so the two improve together.

  5. 5. Manus

    Manus

    Good for
    Hands-off, outcome-first jobs: give a goal and it plans and executes in its own cloud computer, returning built sites, decks, videos, wide parallel research or small games. Twenty tasks run concurrently on Pro, sites deploy directly, and automations trigger on events you define.
    How it works
    An autonomous agent in a cloud virtual machine it controls, with a purchasable always-on cloud computer and the Cue app for personal agents.
    Runs on
    Its own Cascade agent harness (Manus 2.0, 2026-09-28), and the core model is not disclosed. Claude and fine-tuned Qwen were named in March 2025, not since.
    Price
    Free plan with 300 daily refresh credits and one concurrent task, Pro from $20 (4,000 credits) or $40 (8,000) a month, Team from $20 a seat.
    Choose it when
    The deliverable matters more than watching each step: media-heavy builds, site prototypes and parallel research you review once at the end.

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

  6. 6. Gemini Task

    Google

    Good for
    Personal errands and research across a Google account: inbox cleanups, price monitoring, cited research, file creation and browsing that shops, books or reserves, on schedules or triggers. Up to 15 tasks run at once, and results arrive as PDFs, documents, spreadsheets or decks. Renamed from Spark in October 2026.
    How it works
    A personal agent inside the Gemini app, with a remote browser in the cloud or Chrome auto browse on desktop. Cloud tasks run on while devices are closed.
    Runs on
    Google’s own Gemini 3.5 on the Antigravity harness at launch, with Gemini 4 Argon released 2026-09-30 and reaching the Gemini app in October 2026.
    Price
    Needs a personal Google account and Google AI Pro ($19.99) or Ultra ($99.99 or $199.99). Excludes work and school accounts, EEA, UK, Switzerland and Nigeria.
    Choose it when
    Your work and life sit in Gmail, Docs and Chrome, and you want personal tasks handled from a $19.99 plan.

    Sources: Manage tasks in Gemini Apps (Google help), Gemini plans and subscriptions

  7. 7. Genspark

    MainFunc Inc.

    Good for
    Turning one prompt into a deliverable fast: slides, sheets, docs, websites, dashboards, images, video, music and even phone calls placed for you, plus Claw, an agent on its own always-on cloud computer you message through WhatsApp, Teams, Slack or Telegram. SecondBrain memory carries context across jobs.
    How it works
    A Super Agent workspace that plans and executes on the web and mobile, with a Chrome sidebar and Claw running on a dedicated cloud computer.
    Runs on
    Third-party models in a mixture-of-agents design: Genspark says it orchestrates 70+, names OpenAI, Anthropic, AWS and Microsoft partnerships, none of its own.
    Price
    Free tier of about 100 credits a day up to a lifetime allowance, Plus $24.99 a month (10,000 credits), Pro $249.99 (125,000 credits), from secondary captures.
    Choose it when
    You want fast decks, visual production or phone-call agents, and a free tier to try it first.

    Sources: Genspark membership plans (help center), Genspark Workspace 6.0 (press)

  8. 8. Grok Bot

    SpaceXSI (formerly xAI)

    Good for
    Standing routine operations run by named Bots, each with its own role: CRM upkeep, invoice matching, outreach follow-through and vendor portals that have no API. Bots sign into apps and websites, learn a workflow you demonstrate once, share one cloud computer per member and wait for approval on consequential actions.
    How it works
    Personified role agents: you create Bots with names and jobs and message them like colleagues on desktop, mobile, X and Slack, from a shared cloud computer.
    Runs on
    SpaceXSI’s own Grok 4.x models plus undisclosed third-party models chosen automatically. There is no model picker, and you cannot see which model ran.
    Price
    Included with The X Subscription Premium ($30 a month) and above, not with Plus, or with a legacy paid Cursor plan. Usage draws from a shared weekly pool.
    Choose it when
    The job is repetitive operations in apps without an API, and teaching a workflow once by showing it beats writing a specification.

    Sources: Grok Bot on The X Subscription (docs), Grok Bot FAQ (docs)

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 Claude Cowork being discontinued?

No, it is being folded into Claude. When a conversation needs file work, Claude brings Cowork in automatically, and the unified chat-and-Cowork home is rolling out to Pro, Max and Ultra plans. Cowork runs on paid plans from Pro, and Ultra has no 5-hour limit for verified accounts.

Do I have to give up frontier models if I choose Ultramarine Work?

No, Advanced runs a frontier model for the whole task, on Pro and Max, and Assisted uses one as Ultramarine’s advisor, on Plus, Pro and Max. Standard, the default in every mode, runs our own Ultramarine 1 Flash. You choose the way of working per task, so a valuation can run one way and a deck another.

How is a Work mode different from a Cowork plugin?

A Cowork plugin is an extension you enable in a general agent, and it can bundle several skills, connectors and subagents. Anthropic’s investment banking plugin includes skills for CIMs, teasers, process letters, merger models and buyer lists. A Work mode configures the agent for a whole practice: who the work is for, its recurring deliverables and a curated set of skills, 14 for Investment Banking, and it stays selected for the workstream while the agent picks the skill each step needs.

What does Ultramarine Work cost, and is there a free option?

Plans start at $5 a month, with Plus at $20, Pro at $100 and Max at $200. Starter and Plus begin with 7 days free, and there is no free tier. Every plan includes the browser and desktop apps, every professional mode and a monthly allowance of credits, and you can top credits up in 5,000 blocks for $10 when a job needs more.

Which alternative works inside Microsoft 365 under IT control?

Copilot Cowork is the entry built for that. It works across your tenant’s Outlook, Teams, Word, Excel and PowerPoint files with Purview compliance, audit logs and admin spending policies, once an admin enables the Copilot license and assigns usage-based billing. Model choice spans GPT, Claude and Grok families, unless an EU Data Boundary tenant limits it to GPT models.

Where do my files get processed when a task runs?

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