Paper Instruments

ChatGPT alternatives for professional work

A search for ChatGPT alternatives usually wants one of two things: a better assistant for chat and drafting, or an agent that finishes real work in files. Eight options follow, compared on shape, models, price and free options, with an example marketing assignment shown in full.

Checked on

At a glance

ProductBest forAvailable onStarts atFree access
ChatGPTGeneral chat plus the Work agent for file deliverablesWeb, mobile, macOS and Windows desktop appFree, Go $8, Plus $20 a monthFree tier with limits, Work limited on desktop
ClaudeCareful writing, code and the Cowork folder agentWeb, desktop and mobile appsPro $20 a month ($17 annual)Free tier, Sonnet and Haiku, window limits
Google GeminiGoogle ecosystem work and free Deep ResearchGemini app on web, Android, iOS and MacAI Plus $4.99, AI Pro $19.99Free tier, Flash-Lite model, Deep Research, 15 GB
Microsoft CopilotMicrosoft 365 tenant work under admin governanceWindows, Mac, Microsoft 365 apps, webM365 Personal $9.99, Copilot Business $21 listLimited free consumer Copilot
PerplexityCited research and Computer task runsWeb, mobile, desktop apps and Comet browserPro $20 a month, Max $200Free basic searches, some limited Computer turns
Ultramarine WorkProfessional work in configured modes, files you reviewWeb plus desktop app for Mac and Windows$5 a month (Starter)7 days free on Starter and Plus
ManusUnattended runs on a cloud computer with schedulesWeb, desktop and mobile apps, on its cloud computerPro from $20 a month300 daily credits, one concurrent task
DeepSeekLow-cost API, open weights, developer-built agentsWeb, mobile app, API, your own stackFree app, API billed per tokenFree chat app
Vibe (formerly Le Chat)Connectors around Work, Code and Chat modesWeb and mobile apps, CLIPro $14.99 a monthFree tier, limited messages and searches
GensparkOne-prompt decks, sheets and sitesWeb, iOS and Android appsPlus $24.99 a month100 credits daily up to a lifetime allowance

What people use ChatGPT for

ChatGPT is OpenAI’s assistant for drafting, summarizing, data analysis and image generation. Its Work agent handles multi-step jobs that end in files, across connected apps for Slack, Teams, Drive and Salesforce. Work runs in the macOS and Windows desktop app on every plan.

ChatGPT alternatives

  1. 1. Claude

    Anthropic

    Good for
    Claude is Anthropic’s assistant for writing, code and long documents. Cowork, now part of Claude, works across files, folders and connectors and runs tasks on a schedule. The marketplace lists 38 plugins, including role plugins for finance, legal, sales and marketing, built as plain files a team can edit.
    How it works
    A chat assistant first. Cowork adds a general agent over local folders, connected apps and a built-in browser, on desktop, web and mobile.
    Runs on
    Anthropic’s own models: Claude Opus 5.5, Sonnet 5.5, Haiku and Fable 5.1. Cowork runs on the Claude Code harness.
    Price
    Free tier: Sonnet and Haiku in 5-hour and weekly windows. Pro $20 ($17 annual), Max $100, and Ultra $200 with no 5-hour limit for verified accounts.
    Choose it when
    The work lives in documents and code, or your team wants its process in plugins it can read and edit as plain files, with scheduled tasks that run while the laptop is closed.

    Sources: Claude pricing and plans, Claude Cowork support guide

  2. 2. Google Gemini

    Google

    Good for
    Gemini spans the web, Android and Workspace files. Deep Research builds long sourced reports, Canvas drafts and revises documents, Gems store reusable instructions, and the free tier carries the Flash-Lite model, Live, Canvas, Gems and 15 GB of storage.
    How it works
    A chat assistant with a Task agent (formerly Spark) that runs multi-step work on Google’s cloud, even with the laptop closed, up to 15 tasks at once.
    Runs on
    Google’s own models. Task launched on Gemini 3.5 in May 2026. Gemini 4 Argon arrived 2026-09-30, with app availability announced 2026-10-13.
    Price
    Free tier: Flash-Lite model, Deep Research, Canvas, Gems, Live and 15 GB. AI Plus $4.99, AI Pro $19.99, Ultra from $99.99. Task requires AI Pro or Ultra.
    Choose it when
    The work sits in Gmail, Docs and the rest of Google, a free tier has to carry real drafting and research, or a delegated task should keep running on Google’s cloud. Task needs a personal Google account, which excludes managed Workspace logins.

    Sources: Gemini subscriptions and plans, Google Task availability FAQ

  3. 3. Microsoft Copilot

    Microsoft

    Good for
    Copilot sits inside Windows and the Microsoft 365 apps, and Copilot Cowork works in tenant context: it edits Word, Excel and PowerPoint files, sends email, posts in Teams and schedules tasks, pausing for approval before consequential actions. Seven role plugins ship alongside hundreds of partner plugins.
    How it works
    An assistant inside Microsoft 365, plus Cowork, an agent working inside the tenant’s files and admin-set policies, in preview for personal accounts.
    Runs on
    Picker updated 2026-11-05: GPT 6.1 Sol, GPT 6 Astra, GPT 5.6 Terra, GPT 5.5, Claude Opus 5.5, Sonnet 5.5, Fable 5.1, Grok 4.7. EU: GPT only.
    Price
    Consumer Copilot is free with limits. Cowork needs a Copilot license, Business $21 a user list ($18 promo to 2026-12-31), Enterprise $30, plus Copilot Credits.
    Choose it when
    You work in a Microsoft 365 tenant and want the agent inside the files and mailboxes IT administers, with Purview audit trails, data residency and admin spend policies.

    Sources: Microsoft Copilot pricing, Copilot overview and licensing, Cowork model list

  4. 4. Perplexity

    Perplexity

    Good for
    Perplexity answers with citations, so daily use is market scans, sourcing and evidence tables. Computer extends it to agentic tasks that end in files, running sub-agents in cloud sandboxes, and per-thread credit reporting shows what each run spent.
    How it works
    A chat app built on web search, with Computer, a general-purpose digital worker that also runs on your own Mac through Personal Computer.
    Runs on
    Third-party models on Perplexity’s harness. Computer launched on Claude Opus 4.6 (February 2026), and secondary sources report GPT-5.6 models.
    Price
    Free basic searches and some limited Computer turns. Pro $20 (Education $10) adds a weekly Computer allowance, still rolling out. Max $200, 10,000 credits.
    Choose it when
    Most of the work is finding, checking and synthesizing sources, and you want agent use metered in credits with published per-task examples.

    Sources: How credits work on Perplexity, Perplexity on AI costs, Which Perplexity plan is right for you

  5. 5. Ultramarine Work

    Paper Instruments

    Ultramarine Work fits people who use ChatGPT to produce professional deliverables, such as campaign plans, models, memos and decks, and want the agent set up for their practice. A marketer in its Marketing mode gets 13 domain skills, from market sizing and Van Westendorp pricing research to claims review, consent checks and A/B test design. If the need is mostly chat and drafting, the free tiers in the table are likely enough.

    Standard runs Ultramarine 1 Flash, our model post-trained on GLM-5.3-Flash for knowledge work. Assisted adds a frontier model as an advisor, Advanced hands the whole task to a frontier model, and you choose the way of working per task. Paper Instruments writes the modes and trains the model, so a skill such as the Van Westendorp readout can improve together with the model and the code that runs it. Over 130 professional modes carry the depth, the agent loads whichever skill a step calls for, and over 50 connectors bring sources in. You review every file it touched.

  6. 6. Manus

    Manus

    Good for
    Manus runs jobs unattended on its own cloud computer: long research pulls, site checks, batch file builds and scheduled automations. Manus 2.0 added the Cascade harness, Manus Studio and event-triggered automations. The free plan allows one concurrent task and two schedules.
    How it works
    A cloud-computer agent. You set the goal, watch it work in a browser view and collect the files it builds on its virtual machine.
    Runs on
    The Cascade harness, in place since Manus 2.0 (2026-09-28). Manus does not name its core model. A March 2025 report named Claude plus Qwen models.
    Price
    Free plan: 300 daily refresh credits, one concurrent task and two automations. Pro starts at $20 a month for 4,000 credits, or $40 for 8,000.
    Choose it when
    The job is self-contained and can run unattended, such as an overnight site check or a scheduled report build.

    Sources: Manus membership pricing, Manus 2.0 announcement

  7. 7. DeepSeek

    DeepSeek

    Good for
    DeepSeek publishes open weights and bills its API per token, so its models turn up inside developer-built agents, and the chat app is free. The site features DeepSeek-V4-Pro with agent capabilities, and DeepSeek-V4.1-Flash-1017 is open-sourced on Hugging Face.
    How it works
    A free chat app, an OpenAI-format and Anthropic-format API that drops into coding agents and your own stack.
    Runs on
    DeepSeek’s own models: deepseek-flash serving DeepSeek-V4.1-Flash-1017 and deepseek-v4-pro, both with thinking modes and 1M-token context.
    Price
    The app is free and no consumer subscription was found. The API bills per token with off-peak discounts, from $0.15 per million input tokens on cache misses.
    Choose it when
    You want to run a model yourself, or rent it by the token, and wire it into your own tools.

    Sources: DeepSeek home and app, DeepSeek models and pricing

  8. 8. Vibe (formerly Le Chat)

    Mistral AI

    Good for
    Vibe (formerly Le Chat), the name Le Chat took in August 2026, handles chat, coding sessions and delegated multi-step work across Work, Code and Chat modes, carries more than 100 connectors, and its free tier includes basic image generation.
    How it works
    A chat app from the French AI company with Work, Code and Chat modes, many connectors and a command-line coding tool.
    Runs on
    Mistral’s own models.
    Price
    Free tier with limited messages, searches and coding sessions. Pro $14.99 a month ($5.99 for students), Team $24.99 a user.
    Choose it when
    You want a European vendor, a $5.99 student plan, or a chat assistant with many connectors.

    Sources: Mistral pricing, Le Chat is now Vibe

  9. 9. Genspark

    MainFunc Inc.

    Good for
    Genspark’s Super Agent turns one prompt into decks, sheets, sites and phone calls, AgentBase assembles small no-code agents, and Claw runs on a dedicated always-on cloud computer. A third-party test priced a 10-slide deck at about $0.61 in Plus credits.
    How it works
    A task agent with a workspace platform around it, including AgentBase and Claw, under Genspark’s orchestration layer.
    Runs on
    Orchestrated third-party models, 70-plus, routed by Genspark’s agent layer.
    Price
    Free tier: 100 credits a day up to a lifetime allowance, after which refills stop. Plus $24.99 a month ($19.99 annual), Pro $249.99.
    Choose it when
    You want quick one-prompt deliverables, especially slides and visual work. The free daily credits stop at a lifetime allowance, so treat them as a trial.

    Sources: Genspark membership plans, Genspark Plus and Pro price capture

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

Are there free alternatives to ChatGPT that still do real work?

Yes, and each free tier limits use differently. Claude gives Sonnet and Haiku in 5-hour and weekly windows, Gemini pairs the Flash-Lite model with Deep Research, Canvas and Gems, Microsoft Copilot’s consumer version is free with limits, Perplexity offers basic searches free, DeepSeek’s app is free, Vibe limits messages and searches, Manus refreshes 300 credits a day, and Genspark grants 100 daily credits until a lifetime allowance runs out.

Does ChatGPT’s free plan include Work?

Yes, in a limited form and only on desktop. OpenAI’s pricing table marks Work as Limited (desktop app) for Free and Go, while web and mobile access needs Plus at $20 or above. That is enough for occasional small deliverables under tight limits, not for a working launch file set.

How is Ultramarine Work different from ChatGPT Work?

Both turn long, multi-step tasks into finished files. ChatGPT Work is configured with project instructions and plugins from its app directory. Ultramarine Work starts from a professional mode, one of more than 130, each carrying a practice’s skills and review standards across the whole workstream, so you pick the mode once and the agent picks the method each step needs. It runs our own model, trained for knowledge work, and Advanced hands a task to a frontier model.

Is Ultramarine Work free to try?

For 7 days, on the Starter and Plus plans: a card is needed to start, and billing begins on day 8 unless you cancel. There is no free tier. Plans start at $5 a month for Starter with 3,000 credits, and top-ups are $10 per 5,000 credits. If ongoing zero cost is the deciding factor, the free access described in the alternatives is the honest option.

Which ChatGPT alternative is best for research with sources?

Perplexity is built around cited answers, and its Computer agent turns research into files you can check against the sources. Gemini’s free Deep Research and Claude’s web search also cite their sources.

Which alternative suits a company that lives in Microsoft 365?

Microsoft Copilot Cowork is the fit. The agent works inside tenant files with role and partner plugins, Purview audit trails and admin policies, and it pauses for approval before consequential actions. It requires a Microsoft 365 Copilot license (Business $21 list, $18 promo to 2026-12-31, Enterprise $30) plus metered Copilot Credits.

Sources

Checked on