Liveblocks Pricing in 2026: The Collaboration-Minute Meter That Bills You for Your AI Copilot

Quick answer (July 2026): Liveblocks does not bill by monthly active users. MAU is unlimited on every plan. It bills by collaboration minutes at $0.002 per minute, with 3,000 minutes free, and a minute only counts when two or more participants share a room. The catch that surprises teams: an AI copilot is a participant. The moment your agent joins a document, a "solo" writing session becomes a billable two-party room. That single detail is what turns a $30 plan into a $480 one. Below are the numbers, three worked 30-day bills, and the one lever that fixes it.

The meter, decoded

Most realtime and collaboration vendors bill by seats or monthly active users. Liveblocks made a deliberate choice to do the opposite. Straight off its pricing page (July 2026), monthly active users and monthly active rooms are listed as Unlimited on Free, Pro, and Team. You are not charged for having users. You are charged for how much they collaborate at the same time.

The primary meter is the collaboration minute, priced at $0.002 per minute past your included allowance. Per Liveblocks' own definition, a collaboration minute is counted when two or more people or agents are in the same room, and each participant-minute counts. A single user alone in a room costs nothing.

So the meter reads like this:

• Solo session: one human in a room, any duration, $0.
• Two humans co-editing for 10 minutes: 2 participants times 10 minutes = 20 collaboration minutes = $0.04.
• One human plus one AI copilot for 30 minutes: still two participants, so 60 collaboration minutes = $0.12 per session.

That last line is the whole article. Read on.

Here is the rest of the 2026 rate card, so the worked bills below are reproducible:

• Collaboration minutes: 3,000 included, then $0.002 per minute.
• Realtime data storage updates: 3M included, then $1 per 1M.
• Realtime data stored: 1 GB included, then $0.15 per GB.
• Comments created: 200 included, then $0.01 each.
• Custom notifications: 200 included, then $0.005 each.
• File storage: 512 MB included, then $0.15 per GB.
• Plans: Free ($0), Pro ($30/mo, $30 credits), Team ($600/mo, $600 credits), Enterprise (custom).
• Simultaneous connections per room: 10 (Free), 20 (Pro), 50 (Team), 100 (Enterprise).

Credits are the important word. On Pro you pay $30 and get $30 of usage credit; on Team you pay $600 and get $600 of credit. Overages bill at the same pay-as-you-go rate with no markup. Hold that thought for the Pro-versus-Team section, because it changes the "should I upgrade" math.

Bill #1: the disciplined app that pays $0

Start with the happy case, because it shows why the MAU-free model is genuinely good for a lot of products.

You run an internal B2B tool. You have 5,000 registered users. Real co-editing is light: on a typical day two people overlap in a shared document for about 25 minutes, across 20 working days.

• Co-editing minutes: 2 participants times 25 minutes times 20 days = 1,000 collaboration minutes.
• Included free: 3,000.
• Storage, comments, notifications: comfortably under the free caps.

Monthly bill: $0. Five thousand users, zero dollars. Under a per-MAU model that same app would be paying for every one of those 5,000 seats. This is the meter working in your favor: if your users rarely collaborate at the same instant, you barely pay. The bill tracks concurrency, not headcount.

Keep that mental model, because the next bill inverts it.

Bill #2: the AI copilot that turns $30 into $480

Now a document app with an AI writing copilot, the shape half of 2026's startups are shipping.

• 200 daily active writers.
• Each spends about 30 focused minutes a day in a document, with the copilot present in the room so it can see edits and respond in context.
• 20 working days.

The founder's mental math: "200 users, Pro is $30 a month, done." Here is the meter's math instead. Every writing session looks solo, but the copilot is the second participant, so each session is a two-party room:

• Participants per session: 1 human + 1 agent = 2.
• Collaboration minutes per user-day: 2 times 30 = 60.
• Monthly: 60 times 200 users times 20 days = 240,000 collaboration minutes.
• Cost: 240,000 times $0.002 = $480.

Pro includes $30 of credit, which covers the first 15,000 minutes, and the rest bills at the same rate. Monthly bill: about $480. Sixteen times the sticker price, and not one extra human was added.

The uncomfortable part: if you removed the copilot, every session would be a lone human in a room, and collaboration minutes would be $0. The entire $480 is manufactured by agent presence. The agent is not a rounding error on your bill. On this architecture, the agent is the bill.

The one-lever fix

You do not need to drop the copilot. You need to stop leaving it parked in the room.

Lever: join the agent to the room only while it is actively generating, then disconnect it. Most copilots sit idle 90% of a session waiting for the next prompt. If the agent is present for roughly 2 minutes of real generation per 30-minute session instead of all 30:

• Human stays solo the rest of the time: $0.
• Agent presence: 2 minutes times 200 users times 20 days = 8,000 minutes.
• Cost: 8,000 times $0.002 = $16.

That is $480 down to $16, a $464 per month cut, with the live agent presence preserved during the moments users actually see it.

If you want the number even lower, go one step further and keep the AI out of the realtime room entirely: run generation server-side and write results back through Liveblocks Storage via the REST API. The human stays solo, collaboration minutes go to zero, and you pay only the storage-update meter (which, at a few million updates a month, is often still inside the free 3M). Bill: back near the $30 Pro floor.

The honest tradeoff: server-side generation loses the live "watch the agent type" cursor UX. If that presence is a core part of your product, pay for it deliberately. If it is decorative, it is the most expensive decoration on your invoice.

Bill #3: the idle-tab drain

The third trap is quieter and hits multiplayer rooms, not copilots.

You run a live Q&A or review product. A room holds 20 concurrent participants (the Pro cap), sessions run 75 minutes, and you host about 40 sessions a month.

• Active minutes: 20 times 75 times 40 = 60,000 collaboration minutes = $120/mo. Fair enough, that is the product working.

But Liveblocks bills on presence, not activity, and people do not close tabs. Say each participant leaves the room open an extra 40 minutes after the session ends, half-reading, tab in the background:

• Idle minutes: 20 times 40 times 40 = 32,000 collaboration minutes = $64/mo of pure waste, roughly a third of the bill, for rooms where nothing is happening.

Lever: disconnect on tab blur or after an idle timeout. Call the room's leave method when the tab loses focus or a user has been inactive for a few minutes, and tighten the lost-connection timeout. You stop paying for ghosts. That is $64 a month here, and it scales linearly with every idle tab you tolerate.

Pro versus Team is not a volume discount

A natural assumption when the bill climbs: "jump to Team for cheaper usage." Check the credits.

• Pro: $30 for $30 of credit.
• Team: $600 for $600 of credit.

That is the same $1-of-credit-per-$1-paid on both. Team is not a bulk discount on collaboration minutes. What Team actually buys you is limits and features: 50 simultaneous connections per room instead of 20, 50 MB of storage per room, longer version history, SAML SSO, and a private support channel. Upgrade to Team when you hit the 20-connection room ceiling or you need SSO, not because you expect your per-minute rate to fall. It will not.

This is worth saying plainly because it is the opposite of how most infrastructure pricing works, where the next tier is a discount. Here the tiers are feature gates. Your unit cost is flat from your first paid minute to your millionth.

When self-hosting the socket wins

At $0.002 per minute, a small always-on server breaks even fast on paper. A $25/mo box covers 12,500 collaboration minutes, so once your minutes bill clears roughly $100 to $200 a month, running your own Yjs sync layer on a y-websocket server looks cheaper per minute.

The catch is that it is not the same product. Yjs gives you the CRDT sync engine and nothing else. You lose Comments, Notifications, the AI Copilot plumbing, presence webhooks, and the dashboard, all of which Liveblocks bundles into that per-minute price. So the honest rule:

• Self-host the socket when raw realtime sync is essentially all you need, and collaboration minutes are your dominant line item.
• Stay managed when comments, notifications, or copilots carry real product weight, because rebuilding those yourself costs far more than the meter ever will.

If you want the full build-versus-buy framing beyond this one vendor, we walked through the general crossover in when DIY beats SaaS at scale, and the same reasoning we applied to the per-message meter in our Ably pricing teardown maps cleanly onto collaboration minutes.

The pattern

Every one of these bills comes from the same root fact: Liveblocks prices simultaneous collaboration, not users. That makes it cheap when people work alone and expensive when they work together, and in 2026 the fastest way to make people "work together" without noticing is to seat an AI agent in the room next to them.

Model the agent as a participant before you ship, not after the invoice arrives.

Math check: one AI copilot, present for a full 30-minute session across 200 daily writers, costs about $480 a month at $0.002 per collaboration minute. Present only during the 2 minutes it actually generates, it costs about $16. Same feature, same users, a $464 difference decided entirely by when you call connect and disconnect.