Renue By Science · 26 August 2026

Company AI access: what to buy and why

Costed against verified vendor pricing and our own measured usage. Prices USD.

Buy team seats, not individual subscriptions. On Grok and ChatGPT the business seat costs exactly the same as the personal subscription and adds central billing, SSO and offboarding. There is no premium for doing it properly.

The one real exception is Claude Max 20x, which has no team equivalent and is the cheapest usage per dollar of any Claude plan. That is for heavy agent and coding work only, not for seats.

Grok Bot does not come with Grok Business at any price. The cheapest way to put a team on it is a Cursor Teams seat at $40 a head, against $100 for the cheapest individual plan that carries it. Gemini is already paid for and Higgsfield stays for video.

Correcting a common assumption. A company account is not billed at API rates. Claude Team, ChatGPT Business and Grok Business are all flat per-seat with usage included. Only Anthropic Enterprise bills every token at API rates, and it needs 20 seats minimum.source That distinction is what makes team plans viable here.

What a team seat buys, and what it costs

The whole individual-versus-team question comes down to one comparison: what do you give up by buying the personal plan, and what do you pay to avoid giving it up?

Sources: x.ai/pricing · learn.chatgpt.com/docs/pricing · support.claude.com - Team plan
VendorPersonalTeam seatDifference
GrokSuperGrok $30Business $30Identical price
ChatGPTPlus $20Business $20Identical price
ClaudePro $17Standard $20$3 more, for 25% more usage

So the governance is free. What it gets us:

Grok Bot: how access actually works

Worth being precise about, because it is easy to assume the Grok subscription you know carries it. It does not.

Source: the "Compare features across plans" grid at x.ai/pricing, Grok Bot row. Read directly from the page markup - tick on SuperGrok Plus and Heavy, dash on every other column including Business and Enterprise.
Plan$/monthGrok Bot
Grok free, and SuperGrok Lite0 / lowNo
SuperGrok30No
SuperGrok Plus100Yes
SuperGrok Heavy300Yes
Grok Business30/userNo
Grok EnterprisesalesNo

Grok Business is the seat-managed version of Grok chat and Build. It gives us the admin controls and central billing, and it is the right standard for everyday Grok use. It does not carry Grok Bot, and neither does Grok Enterprise. That is read straight off xAI's own plan comparison, not inferred.verify

Plain SuperGrok at $30 does not carry it either - the tick starts at SuperGrok Plus, $100.

Grok Bot is sold on its own set of plans:

Source: the Pricing block on x.ai/bot, which lists exactly these plans and closes with "Already on an eligible Cursor, SuperGrok, or Teams plan? Grok Bot is included."
Route to Grok Bot$/monthWhat it adds
Cursor Pro+60Cheapest individual route
SuperGrok Plus100Bot plus the Grok chat upgrade
Cursor Ultra200Extended limits on AI tokens
SuperGrok Heavy300Highest usage, fastest speed
Cursor Teams Standard40/seatCentral team billing, shared usage analytics, team marketplace for skills and plugins
Cursor Teams Premium120/seatAdds SAML/OIDC SSO

This is the clearest case in the whole analysis for buying team rather than individual. A Cursor Teams Standard seat is $40 against $60 for the cheapest individual plan carrying Bot, and $100 for SuperGrok Plus. It is 60% cheaper per head than the individual route most people would reach for, and it is the only route that comes with central billing.

It also reprices what we do now. One Cursor Ultra at $200 covers one person. The same $200 as Cursor Teams Standard covers five.

One thing to confirm before buying: whether a Cursor Teams seat also carries grok.com chat, or only Grok Bot. If it does, $40 a head covers everything. If it does not, the full standard is Grok Business at $30 plus Cursor Teams at $40, which is $70 a head - still well under the $100 individual route that has no team administration at all.

Where individual subscriptions genuinely win

Exactly one place. Anthropic publishes usage as a multiple of the Pro plan, which makes the plans directly comparable:

Sources: seat prices and the 1.25x / 6.25x multipliers from support.claude.com - What is the Team plan?; Pro and Max prices from claude.com/pricing. The Max 20x price is also confirmed against our own Anthropic invoice.
Claude plan$/monthUsage vs ProCost per 1x unit
Pro171.00x$17.00
Team standard201.25x$16.00
Team premium1006.25x$16.00
Max 5x1005x$20.00
Max 20x20020x$10.00

Max 20x is the cheapest usage per dollar by a factor of 1.6, and a Claude Team seat tops out at 6.25x. Two premium seats at $200 would buy 12.5x against Max 20x's 20x, and one person cannot hold two seats anyway. For genuinely heavy agent work there is no team product that competes.

Max 5x is strictly dominated and should never be bought: a Team premium seat is the same $100 for 25% more usage plus every admin control.

The rule this gives us

A Claude Team premium seat is 6.25x Pro for $100. Individual Max 20x is 20x Pro for $200.

Each person's actual consumption is visible to them in Claude under Settings, then Usage. That is a measurement, not a guess, and it should decide each case.

Applying it: the people already on individual Claude Max

Assume four people hold personal Claude Max accounts today. That is $800 a month, on personal cards, invisible to Bryan, and unrecoverable when someone leaves.

ScenarioMonthlyResult
4 individual Max 20x, as today$80080x total usage, no admin visibility
1 Max 20x + 3 Team premium seats$500Same work if the three sit under 6.25x
Saving$300Plus central billing on three of the four

Measure the three before moving them. Anyone above 6.25x stays on Max, and that is the correct outcome, not a failure of the plan.

Why not just use the API

This is the assumption worth killing with data. I measured the actual Claude Code token consumption on our machines over the last 30 days and priced it at Claude's published API rates:

Priced against the published per-million-token rates at platform.claude.com/docs - Pricing, including the separate cache-write and cache-read multipliers. Method below.
MachineBillable callsAt API list rates
RBS VPS10,464$3,138
Rod's desktop2,061$874
RBS total12,525$4,012

The subscriptions covering that usage cost $217 a month. That is a ratio of 18 to 1.

Read it as consumption value rather than a bill we dodged, because at metered rates we would simply work differently. But it settles the direction: agent and coding workloads belong on subscriptions, and moving them to metered API would be the single most expensive decision available to us.

It also rules out Anthropic Enterprise, where the seat fee covers access only and every token bills at API rates on top. At our consumption that is seat fees plus roughly $4,000 a month, before the 20-seat minimum is considered.

What to buy

LineSeats$/monthPurpose
Cursor Teams Standard @ $4010400Grok Bot for everyone, centrally billed.
Grok Business @ $3010300Grok chat as the company standard. Drop this line if a Teams seat already covers chat.
Claude Max 20x (individual)1200Heavy agent work. No team equivalent exists.
Claude Team premium @ $1003300The other Max holders, once measured.
Gemini via Workspaceall0Already paid for. Free second opinion.
HiggsfieldsharedexistingAI video. Unchanged.
ChatGPT Business @ $20on request-Only where someone needs it specifically.
Run rate$1,200$900 if Teams covers chat

Against what we already pay

Identified today$/month
Claude Max 20x200
3 further individual Claude Max600
Claude Pro17
ChatGPT Plus on a company card since Dec 202520
Cursor Ultra, one person, for Grok Bot200
Grok via x.com, and Higgsfieldnot quantified
Identified$1,037

$1,037 currently covers about five people, on personal cards, with no admin visibility and nothing recoverable at offboarding. Only one of them has Grok Bot.

$1,200 puts ten people on Grok Bot and Grok chat, three under managed Claude seats, and everything on one bill - or $900 if a Cursor Teams seat turns out to cover grok.com chat as well.

The single largest saving available is the Cursor line. One Ultra seat at $200 covers one person; the same money as Cursor Teams Standard covers five.

Sequence

The questions from the thread

Do we have a Grok, Claude, or ChatGPT account? Maria, Elsa, Jason

Not a shared one. What exists today is a scatter of individual subscriptions, some on company cards and some on personal ones. Under this proposal every one of you gets a Grok Business account on your @renuebyscience.com address, and Gemini is already yours through the Workspace.

Is Grok, rather than Claude, the best option for page audits? Maria

Grok is what that review was run on and it is the proposed company standard, so it is the default. Two things matter more than which model:

  • Run it twice. Gemini costs nothing extra, so a second pass on a different model is free and catches different things.
  • Save the rubric as a shared project, so every audit asks the same questions instead of depending on who wrote the prompt.

Claude stays reserved for agent and coding work, where the usage economics are completely different.

Should AI audits be a standard step for every page we build? Maria, Elsa

Yes, in the shape Elsa described: an AI first pass to flag friction points and inconsistencies, then human validation of the findings. The shared project is what makes it repeatable - and it lets Elsa record the places where we have diverged from expected UX patterns deliberately, so the audit stops re-flagging them every time.

That also fixes the problem Jason hit on this one, where placeholder copy generated findings that were not applicable. Run it when the page is close to final, not before.

Can it plug into our systems and assets? Jason

Gemini already connects to the RBS Google Drive - use your RBS account, as Ramon said. Grok Business adds connectors and shared projects on top. Between them they cover what you are currently doing on your personal Claude and ChatGPT.

How the usage figure was produced

Parsed every Claude Code session log on each machine, kept messages from the last 30 days, and deduplicated on the API request ID - which matters, because session forking and compaction write the same message into several files and 8,954 duplicate records were discarded. Each remaining call was priced per model against Anthropic's published per-million-token rates, including the separate five-minute and one-hour cache-write multipliers and the cache-read rate.

Multi-iteration records were checked as a possible undercount and are immaterial: 7 records out of 120,339. The figure is conservative.

Sources

Every price and capability claim on this page comes from one of these. All checked 26 August 2026.