Best AI Assistant for Work in 2026: Priced, Not Tested

Four of the lists ranking for AI assistants at work claim a hands-on test. A two-month test. Fourteen tools tested. A whole category tested. I have no way of checking whether any of those tests happened, and neither do you.

So here is a plainer promise. On 21 September 2026 I opened the pricing pages for Claude, Notion and Google Workspace and wrote down what they said, and requested OpenAI's. Two more rows are dated earlier: Microsoft 365 Copilot verified 19 September, Lindy from a capture dated 15 September. That is the whole method. No bake-off, no trial account, no scoring rubric I invented on a Tuesday.

Two of those pages did something worth reporting. OpenAI's refused my request outright. Google's served me Indian pricing on a promotion that expires on a date the page prints itself. No competing page I have seen carries the currency warning, and both facts change what you should expect to pay.

My own product is in here. Its row sits last, and on the price column it does not win. I would rather you trusted the rest of the table.

How I compared them

Every tool got the same four questions, and all four are answerable from public documentation.

• What does the vendor's own page charge, on a stated date? The vendor's page itself, with the two earlier-dated rows called out in the table below. A review site's summary of a price is a second-hand price and it ages badly.
• Who does the work once you have paid? Most of these answer questions and hand the answer back to you. One of them prints task handoff on the plan line. That is a meaningfully different product and the price does not always reflect it.
• Where does the AI actually switch on? Several vendors put a trial on the cheap tier and the real feature two tiers up. The price worth comparing is the one where the thing you want exists.
• What would the page not tell me? A refused request and an unrequested currency are both findings. I wrote them down.

One thing I did not do: run these tools against a workload. This is a pricing and documentation review, dated row by row in the table below. I am not claiming a test, and every number below can be checked in about ninety seconds.

Quick comparison

Anthropic, Notion and Google Workspace prices are what those pages showed on 21 September 2026. Microsoft 365 Copilot was verified on 19 September 2026 and not re-read. Lindy is from a stored capture dated 15 September 2026. OpenAI's pricing page refused the request on 21 September 2026, so that cell carries the refusal instead of a number. Gravity's own pricing is as of September 2026. Nothing here has been converted between currencies.

Two rows are not prices, and that is the finding

A price table is only as good as the reads behind it, so the two reads that did not go to plan are worth more space than the ones that did.

OpenAI's pricing page returned HTTP 403 to my request on 21 September 2026

I asked for openai.com/chatgpt/pricing/ and the server refused. HTTP 403 means the request was understood and declined, which is a different situation from a page being down or a URL being wrong. So this post carries no ChatGPT price, no plan name and no capability claim.

Filling that cell from memory is exactly how a comparison post goes quietly wrong six weeks later. The most obvious tool in the category therefore appears as a blank row with a reason attached. You can reproduce the check yourself: the URL and the date it refused are item four in the Sources list at the bottom of this page. If it loads for you, you have learned something about the difference between your network and mine.

Google Workspace served me Indian pricing inside a window the page itself dates

workspace.google.com/pricing came back in ₹ rather than dollars, without being asked. I have not converted it, because a converted price is an invented price. Here is exactly what the page showed, with the tab set to Annual and labelled "Save 16% with 1 year commitment":

• Base, as served: ₹99.00 per user a month, struck through, with ₹49.50 printed beside it.
• The window, printed on the page: "Oct 5, 2026 to Jan 5, 2027, 50% off, for all users". From 5 January 2027 the same page shows ₹99.00 again for all users.
• Starter, as served: ₹270 per user a month, described on the page as everything in Base plus more.

Two warnings fall out of that. The headline figure is a promotional rate with a printed end date, so anyone signing in February pays double what I saw on screen. And if you are reading this outside India, your own page will show a different currency and quite possibly a different offer. That is not a rounding error in a price comparison. It is the gap between the number on the screen and the number on the invoice.

1. Claude: the cheapest plan here that names handing off a task

Anthropic's pricing page does something the rest of this set mostly does not. It prints both billing states on the same line. Pro is $17 a month with the annual subscription discount, $200 billed up front, and $20 if you pay monthly. You are never left guessing which of those two numbers you are looking at, which sounds like a small thing until you have read several pricing pages in a row.

The line that matters for work sits on the Pro plan: "Hand off and schedule tasks". At $17 a month that is the most agent-shaped sentence at the lowest price in this comparison. Pro also lists Claude Code, Claude Design, Slides and Docs, Claude Science, Projects, Claude in Chrome and Microsoft 365, and Claude for Microsoft Outlook.

The Free plan is a real one. It covers chat on web, desktop and mobile, searching the web, creating files and running code, memory across conversations, connected apps and tools, and Artifacts. Max starts at $100 a month and lets you choose 5x or 20x more usage than Pro, with higher output limits, early access and priority when traffic is heavy.

Two things on the page deserve a read before you commit. The footnote says usage limits apply, that prices exclude applicable tax, and that price and plans are subject to change at Anthropic's discretion. A banner on the same page on 21 September 2026 read "Claude Cowork is now just Claude. Rolling out to Pro and Max, with more plans to follow." Product names in this category move faster than the prices do, so date your own notes.

Best for

One person who writes, codes and wants a scheduled handoff without starting a seat-licence conversation with procurement.

Pricing

Free at $0. Pro at $17 a month on the annual subscription ($200 billed up front) or $20 billed monthly. Max from $100 a month. Read 21 September 2026.

Free tier

Yes, and a usable one.

Verdict

The lowest price in this table that names task handoff on the plan itself. If the category words are blurring together, the difference between an agent, a chatbot and an assistant is the shorter read.

2. Microsoft 365 Copilot: $18.00 per user, and the price has an expiry date

Microsoft 365 Copilot was $18.00 per user a month paid yearly when I verified it on 19 September 2026, down from $21.00. I have not re-read it since, so treat that row as two days older than the others. If the word copilot is doing real work in your shortlist, how an AI agent differs from a copilot is the separate argument and it changes what you should be shopping for.

The important part is not the discount. It is the expiry. The promotional $18.00 expires on 31 December 2026, and the struck-through $21.00 is what I would budget against after that. On a 40-seat team that gap is $1,440 a year arriving in January, on a renewal nobody diarised. The arithmetic is $3 of difference, 40 seats, 12 months, straight off the two published figures.

So the instruction for this row is short. Write 31 December 2026 next to the Copilot line in whatever sheet you are building, and model the renewal at $21.00 per user a month rather than $18.00.

3. Notion: the agent arrives on the Business plan, not before

Notion's tiers read cleanly until you look at where the AI sits. Free is $0 per member a month. Plus is $10 per member a month and adds basic connections to Slack and Google Drive. Both of those list a trial of Notion AI, and a trial is precisely what the word means.

Business is $20 per member a month and it is where Notion Agent appears. The page describes it as: "Does work for you. Completes complex, multi-step tasks using context from Notion, your connected apps, and the web." Business also carries AI Meeting Notes, Enterprise Search in beta, and SAML SSO.

Free and Plus give you a trial, and the agent starts at Business

This is the most common mispricing in the category and it is worth its own paragraph. Shortlist Notion at $10 because the tier mentions AI, and what you have actually shortlisted is a trial. The agent lives at $20 per member a month, double the figure that got Notion onto your list. Across ten people that is $1,200 a year of difference, on the published per-member rates.

One honesty note. The capture I read did not show whether the page was in its monthly or its annual state, so I am not going to tell you Notion's billing period. The per-member figures above are what the page displayed, and nothing more than that. What Notion's agent does inside a workspace covers the feature side properly.

4. Google Workspace: the suite is the purchase, and the price came in a currency I did not ask for

Workspace is the odd row here, because the page sells the whole stack and never sells you an assistant. Base includes 20 GB of pooled storage per person, custom business email, the full app suite, and video meetings for 100 participants. There is a 14-day free trial. The page I read does not price an assistant as a separate line, so I am not going to describe one.

The prices, as served to this machine, were ₹99.00 per user a month for Base, discounted to ₹49.50, and ₹270 per user a month for Starter, with the tab set to Annual and labelled "Save 16% with 1 year commitment". I covered the currency problem above and will not repeat it, apart from the part that moves your budget: the ₹49.50 is promotional and the page says the window closes on 5 January 2027.

Treat this row as a budget floor. If you need business email and shared docs at all, you are paying for something like Workspace whichever assistant you pick, and the AI decision is a separate line on top of it. Read the promotional dates on your own version of the page before you sign a one-year commitment, because the window I saw runs from 5 October 2026 to 5 January 2027 and yours may carry a different offer entirely.

5. Lindy: priced per person, at the top of this table

Lindy is the most expensive row here and the easiest to read. The tiers are $29.99, $99.99 and $199.99 per user a month, with a 7-day trial. That comes from a stored HTTP 200 capture of lindy.ai/pricing in the pricing-watch archive, dated 15 September 2026, six days ahead of the rest of the table.

Model it per person before you compare the headline. Five people on the entry tier is $149.95 a month on the published rate, more than seven individual Claude Pro subscriptions at $20 each. The arithmetic is five times $29.99 against seven times $20, straight off the two published figures. Whether that is worth it depends on how much work it actually removes, and no pricing page has ever answered that question for anybody.

Best for

A single operator, or a small team buying one per-seat licence each and comfortable with paying by headcount.

Pricing

$29.99, $99.99 and $199.99 per user a month, with a 7-day trial. Captured 15 September 2026.

Free tier

None found in that capture. The trial runs 7 days.

Verdict

The clearest per-seat pricing in this set and the highest entry price. How Lindy's pricing has changed over time tracks the movement, and Gravity and Lindy side by side is the direct comparison.

6. Gravity: my own row, and it does not win this table on price

This is my product, so discount it accordingly. I put it last, and I am going to open with the part that argues against it.

Pro is $5 a month during the alpha, and Max is $20 a month for more agents and usage. If you run out, you buy more usage. Your first agent is free and there is no card. Gravity is in private alpha, so the way in is an application, and the catalogue is short because private alpha means short.

Max at $20 a month is exactly Claude Pro billed monthly and exactly Notion Business per member, and it sits above Microsoft 365 Copilot's current promotional $18.00. Pro at $5 is an alpha price with $5 of monthly usage included, so on a heavier workload Gravity is mid-table, and Claude on the annual subscription undercuts everything else here. I would rather write that sentence than build a column that flatters me.

I put my own row last because the table does not favour it

The column where these products genuinely diverge is the third one, who does the work. Most of the rows above sell an assistant: you prompt, it answers, you decide what happens next. Gravity sells a finished agent that an independent builder has already made and tested. You describe the task in plain words, connect the accounts through an ordinary login, approve anything that touches a customer, and it runs on a schedule.

A run that fails on a platform error, with no usable output, is refunded under the refund policy.

Best for

A recurring work task you want finished on a schedule while you are busy doing something else.

Pricing

First agent free, no card. Then Pro at $5 a month during the alpha, or Max at $20 a month for more agents and usage. Buy more usage if you run out. As of September 2026.

Free tier

Yes. One agent, free.

Verdict

Mid-table on price, and a different shape of product. If a published agent covers your task, none of the priced rows above removes the work the same way. If no agent covers it, one of those rows is your answer and I would rather you took it.

Which one should you choose?

Start from what you already pay for, because it settles about half of this. If your company runs on Microsoft 365, Copilot at $18.00 per user a month paid yearly is the closest thing here to a decision you have already half made, and the only thing to remember is that the promotional rate expires on 31 December 2026 and the struck-through $21.00 is the figure to budget against afterwards.

If your work lives in Notion, Business at $20 per member a month is the plan where the agent actually exists. The $10 Plus tier will not get you there, whatever the feature list implies.

For one person doing knowledge work, Claude Pro is the strongest value in this set at $17 a month on the annual subscription, with task handoff and scheduling printed on the plan itself. The Free plan is also good enough that you can find out before you pay.

Google Workspace is barely competing with the rest of this list. You buy it for business email and shared documents, and the assistant question arrives afterwards as a separate cost.

Lindy is the pick when you are buying one per-seat licence per person and have budgeted for paying by headcount, at $29.99 per user a month on the entry tier.

ChatGPT may well belong on your shortlist. I cannot rank it here, because its pricing page would not serve me a price on the day, and ranking a tool on a number I did not read is how these posts lose their value. The Sources list carries the URL so you can check it yourself.

My own bias, stated plainly: for a work task you repeat every week, I would pick the option where nobody has to prompt anything. That is Gravity: the first agent is free, Pro is $5 a month during the alpha, and Max costs the same $20 a month as two of the rows above.

If none of that decides it, change the question from which tool to which job. What a personal AI assistant actually is and personal AI agents compared for home and side projects both come at the problem from the other end.

FAQ

Sources