The End of the AI Flat Rate? Musk Says Grok Bot Cost Too Much, OpenAI Tightens the Rules. What Changes for You

By Thorsten Meyer | November 2, 2026 | ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

In December 2024, OpenAI launched a $200-a-month ChatGPT plan with “unlimited access” to its best models. One month later, OpenAI’s chief executive Sam Altman posted on X: “insane thing: we are currently losing money on openai pro subscriptions! people use it much more than we expected.”

On Sunday, 1 November 2026, Elon Musk said much the same thing about his own AI agent. A user asked him for a Grok Bot reset because “lots of people say their usage ran out really fast this week.” Musk replied: “The cost of serving Grok Bot was unsustainably high relative to revenue.” In the same week, OpenAI published new limits on how many paid ChatGPT accounts can share one device or one payment card, and announced that ChatGPT and Codex will draw from one shared allowance from December.

Put those together and a question appears: is the all-you-can-eat era of AI subscriptions ending? I read the original posts, the help and pricing pages of OpenAI, xAI and Anthropic, and the reporting around them. My answer: the flat price is staying. What is ending is the idea that a flat price means unlimited use.

A note before we start: Anthropic makes Claude, and Claude helped research and draft this article. Anthropic also limits its subscriptions, so it is part of this story, and I treat its rules the same way as everyone else’s.

Key takeaways

• Musk’s post is real, short and has no numbers. On 1 November at 20:34 UTC he wrote that the cost of serving Grok Bot “was unsustainably high relative to revenue.” He gave no cost or revenue figures and did not say what changed. It is about Grok Bot, xAI’s agent, not Grok chat in general.
• OpenAI closed a loophole. Its help center now says ChatGPT limits how many paid accounts can be used on one device and how many subscriptions one card or device can buy. That covers Go, Plus and the three Pro plans. OpenAI publishes no number.
• One meter instead of several. From December, Plus, Pro and Business plans get one allowance across ChatGPT and Codex, measured in tokens, according to OpenAI’s help center. Unlimited “Instant” chats stay outside that meter.
• More allowance and more metering arrived in the same week. On 29 October OpenAI doubled the Work and Codex allowance on Pro 100 and Pro 500 and cancelled a planned cut on Pro 200.
• Why it is happening: a question no longer has a fixed cost. On OpenAI’s own rate card, the most capable model uses up your allowance about 100 times faster per output token than the cheapest one. Agents that work for minutes or hours use far more than a chat reply.
• For companies: budget AI like cloud, with a seat price plus metered use, and ask vendors for numbers in writing. For users: check the usage page before you pay for an upgrade, a reset or credits.

Products Worth Considering

Bose SoundLink Plus Portable Bluetooth Speaker, Wireless Outdoor Speaker, Up to 20 Hours Battery Life, Waterproof and Dustproof, Black

AmazonView Latest Price

Jackery HomePower 3600 Plus Portable Power Station, 3584Wh, 3600W

AmazonView Latest Price

Bose SoundLink Flex Bluetooth Speaker (2nd Gen), Waterproof, Dustproof, USB‑C, Up to 12 hr Battery, Black

AmazonView Latest Price

As an affiliate, we earn on qualifying purchases.

What Musk actually said, and what he did not

The exchange took two posts. At 20:11 UTC on 1 November, the X user DogecoinNorway asked: “Any chance for a Grok Bot reset, @elonmusk? Lots of people say their usage ran out really fast this week.” Twenty-three minutes later Musk answered with one sentence: “The cost of serving Grok Bot was unsustainably high relative to revenue.”

Grok Bot is xAI’s agent. It does not just answer a question. It takes a task, like research, writing or browsing, and works on it in the background. Today it is part of the X Subscription, which xAI launched on 12 October 2026. According to xAI’s pricing page, Grok Bot is part of the Premium plan at $30 a month, Super at $100 and Ultra at $200, and not part of Free or Plus. Its usage “draws from that seat’s shared weekly usage pool,” the same pool that covers Grok chat and xAI’s models in the coding tool Cursor.

What the post does not say matters as much:

• No numbers. Musk gave no cost, no revenue and no margin. Anyone who puts a figure on Grok Bot’s losses is guessing.
• No new limit. He did not say whether the weekly pool was cut, by how much, or whether users will get a reset or a price change.
• “Was”, not “is”. Several outlets read the past tense as a sign that a cost change already happened. That is a reading, not a statement.

Reports also describe a post from Grok’s official account that blamed “adjusted allocations” on the same costs and mentioned a separate, fixed dashboard fault that showed usage too high for a while. I could not find that post itself, so treat it as reported, not confirmed.

xAI’s own documentation explains why Grok Bot is expensive to run. Its usage is “measured by agent steps and tokens rather than the number of messages you send,” and “how fast you use your usage depends on the kind of work you give Grok Bot and how much effort each task takes.” In other words, a flat monthly price sits on top of a meter that runs at a different speed for every task.

What OpenAI changed in one week

OpenAI made three changes between 29 October and 1 November. They point in different directions, which is why the headlines got confusing.

1. More allowance on the top plans (29 October). Thibault Sottiaux, who posts as Tibo and works on Codex and ChatGPT, wrote on X: “We are doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today.” OpenAI’s release notes are narrower: Pro 100 and Pro 500 get twice their previous Work and Codex allowance, and a planned cut on Pro 200 is cancelled. We covered the details in OpenAI Doubles ChatGPT Pro Limits and Drops the Pro 200 Cut.

2. Limits per card and per device (help page updated in late October). OpenAI’s help article “Understanding card and device limits in ChatGPT” says: “If too many subscriptions are associated with the same payment card or device, ChatGPT may block a new subscription purchase.” ChatGPT also limits how many paid accounts can be used on one device. The key points:

• It applies to ChatGPT Go, Plus, Pro 100, Pro 200 and Pro 500. Free accounts and Business, Enterprise and Edu workspaces do not count.
• OpenAI does not say what the limit is.
• Buying credits or resetting usage does not lift a card or device block.
• A family that shares one card can hit the limit even if everyone uses their own device.
• OpenAI calls the rule part of its “privacy and security protections.”

Who does this hit? Mostly people who switched between several cheap paid accounts on one computer to get more usage, especially for Codex. The Decoder’s reading is that OpenAI is steering those users toward credits and paid resets, which usually cost more than another subscription. That is the outlet’s interpretation, but the help page points the same way: the alternatives it names are credits and usage resets on an account you already have.

3. One shared meter from December (help page updated 31 October). OpenAI’s article “Understanding ChatGPT and Codex usage limits” says that “starting in December,” the usage allowance on Plus, Pro and Business plans “will be shared across ChatGPT and Codex.” Usage will be measured in tokens, the small pieces of text a model reads and writes. More advanced models and complex tasks use more. Remaining usage and reset times will appear under Settings, then Usage. If you reach the limit, you can wait, upgrade or buy credits. BleepingComputer saw OpenAI’s email to subscribers, which says: “Your use of ChatGPT and Codex will draw from the same allowance.” According to the help page and the email, unlimited Instant chats are included in the subscription and do not count toward the shared limit.

OpenAI says it expects most users won’t have to wait for a reset more often than they do now. That may be true. But the direction is clear: the allowance becomes one number, measured in tokens, with paid ways to get more.

What “credits” and “resets” mean

• Credits are a prepaid balance. OpenAI’s help page says your included usage is used first, then the credit balance. Prices can vary by account, region and plan, credits are not API credits, and they expire after 12 months.
• A paid reset gives you your weekly Work and Codex usage back immediately. But it does not add extra usage. According to OpenAI, it “pulls your normal weekly allowance forward” and moves your next weekly reset date. It is offered to selected Plus and Pro accounts, the price is shown only at checkout, and OpenAI generally does not refund it.

Why a flat rate is hard when every question costs something different

A gym can sell a flat rate because one visit costs it roughly the same as the next. A question to an AI does not. Three things make the cost swing.

The model you pick. OpenAI publishes how fast each model uses up a Codex and Work allowance, in credits per million tokens. For output, GPT-6 Luna, its small model, costs 12.5 credits. GPT-6.1 Sol costs 250. GPT-6 Astra, its top model, costs 1,250. That is a factor of 100 between the cheapest and the most capable model. OpenAI’s own estimate for a Plus user shows what that means in practice: in a five-hour window, about 350 to 3,000 messages with Luna, but only about 5 to 45 with Astra.

The way it thinks. Reasoning models write out long chains of intermediate steps before they answer. You don’t always see them, but you pay for them in tokens. Faster modes cost more too: OpenAI’s pricing page says Fast mode uses 2.5 times the included usage and Ultrafast eight times.

The length of the job. An agent that researches, writes code or clicks through websites for an hour makes many model calls in a row. The coding tool Cursor explained this in July 2025, when it moved from counting requests to charging for actual usage: “the hardest requests cost an order of magnitude more than simple ones.” OpenAI made the same move for Codex on 2 April 2026, when it “updated Codex pricing to align with API token usage, instead of per-message pricing.”

A flat price on top of a cost that swings 100 times is a bet. The provider bets that most people use little, so the heavy users are paid for. Agents break that bet, because one user can keep a machine busy all day. That is what Musk’s sentence describes. The research firm SemiAnalysis measured how much API value the big subscriptions hand out, and we looked at its findings in The 5x Is a Subsidy, Not a Price.

A short history of “unlimited”

The pattern is older than this week. Each step below comes from the company’s own announcement or pricing page.

• 1 February 2023: OpenAI launches ChatGPT Plus for $20 a month, with access “even during peak times” and faster answers.
• 5 December 2024: OpenAI launches ChatGPT Pro for $200 a month, with “unlimited access” to its then best models.
• 5 January 2025 (US time): Altman writes that OpenAI is “currently losing money” on Pro because “people use it much more than we expected.”
• 4 July 2025: Cursor apologises for a pricing change that replaced 500 monthly requests with $20 of usage at API rates, and offers refunds.
• 28 July 2025: Anthropic announces weekly limits for Claude Pro and Max from late August, estimating they will affect “less than 5% of subscribers.”
• 2 April 2026: OpenAI moves Codex from per-message to token-based pricing.
• 6 May 2026: Anthropic doubles Claude Code’s five-hour limits after a compute deal. More chips, more allowance.
• 12 October 2026: xAI merges SuperGrok, Cursor and X Premium into the X Subscription, with one weekly usage pool.
• 29 October to 1 November 2026: OpenAI doubles top-plan allowances, limits accounts per card and device, and announces one shared meter from December. Musk calls Grok Bot’s serving cost “unsustainably high.”

Read the list once more and you’ll see that limits move both ways. When a provider gets more computing power or faces a competitor, allowances go up. When demand outruns the chips, they go down. What never changes is that the provider sets the meter.

Anthropic does it too

Anthropic, the company behind the assistant that helped draft this article, runs the same model. Its pricing page lists Pro at $20 a month, Max at $100 with “5x Pro usage every 5 hours and every week,” and Ultra at $200 with “10x Pro usage every week” and no five-hour limit for verified accounts. Ultra used to be called Max 20x. Anthropic’s support page now says: “The Max 20x plan is now called the Ultra plan.” Every paid plan can add usage credits once the included usage runs out. And the fine print is the same as everywhere else: how quickly you hit the limits “depends on the model, the length of your chats, and the size of your files.”

So this is not one company’s problem. Every big AI subscription now has the same three parts: a monthly fee, an allowance that is metered behind the scenes, and a paid way to get more.

Is the flat rate really ending?

Not the price. You still pay $20, $100 or $200 a month, and some things stay unlimited, like OpenAI’s Instant chats. What is ending is the promise that the price covers everything.

A phone plan is the closest comparison. You pay a monthly fee and get a data allowance. Calls may be unlimited, but heavy data use is metered, and when you run out you can buy more or wait for next month. AI subscriptions are moving to that model.

There is a counterargument, and it deserves a fair hearing. In the same week, OpenAI doubled the allowance on two of its Pro plans, and Anthropic has raised limits when it gained computing power. Competition pushes allowances up. But even then, the allowance is a number the provider sets and can change, and the unit is moving from “messages” to “tokens”, which is harder for a normal user to see.

What this means for companies

• Budget AI like cloud, not like software licences. A seat price is only the minimum. Plan for usage on top: credits, overage or API spend. Check monthly what you actually use.
• Ask for numbers in writing. “Generous” and “5x” are not limits. Ask the vendor what a seat includes, in tokens or credits, what happens at the limit and how much notice you get before changes.
• Separate light and heavy users. Most staff need chat. A few run agents all day. Per-seat plans for the many and usage-based plans or the API for the few is often cheaper than top plans for everyone.
• Do not build on account stacking. If a team shares accounts or switches between several cheap ones on one machine, OpenAI’s new device rules can lock people out without warning. Use proper business seats.
• Watch the lock-in. Shared pools across products, like ChatGPT plus Codex or Grok plus Cursor, are convenient, but they make it harder to switch one tool without losing the other. Keep your prompts, data and workflows portable.
• Use the price war. Vendors are competing hard for business customers. We looked at the fine print in The AI Price War.

What this means for users

• Look at your usage page first. ChatGPT, Claude and Grok all show remaining usage and reset times in their settings. Look before you upgrade.
• Pick the model for the job. For simple questions, use the fast, light model. Save the top model for hard problems. On OpenAI’s rate card, that difference is up to 100 times.
• Think twice about paid resets. On ChatGPT, a reset gives you next week’s usage early. It does not add new usage, and OpenAI generally does not refund it.
• Turn off automatic top-ups or set a spending cap. xAI warns that a running task “can go a little past it.”
• One account per person. Families sharing a card can hit OpenAI’s checkout limit even on separate devices. If a purchase fails with “Too many subscriptions associated with this card or device,” that is OpenAI’s rule, not your bank.
• Choose by what you do, not by the multiplier. “10x” sounds big, but if you mostly chat, a cheaper plan may be enough.

Limits of this piece

• No figures from xAI. Musk gave no cost or revenue numbers, and xAI has not published how Grok Bot’s limits changed. The Grok account’s explanation is reported, not seen by me.
• OpenAI’s help center blocked my automated reader. I read the help pages through search-engine copies of the pages and checked them against The Decoder and BleepingComputer. The quotes match across those sources.
• No December date. OpenAI says “starting in December” and gives no day. It also has not said how the 29 October doubling carries over into the shared allowance.
• No published limits. OpenAI does not say how many paid accounts a device may hold, and it shows reset prices only at checkout.
• Neutrality. Anthropic makes the assistant that helped draft this piece. Its plans are described with the same yardstick as OpenAI’s and xAI’s.

Sources

• Elon Musk on X, 1 November 2026: post, replying to DogecoinNorway
• xAI: X Subscription pricing, Introducing The X Subscription (12 October 2026), Grok Bot FAQ
• OpenAI help center: Understanding card and device limits in ChatGPT, Understanding ChatGPT and Codex usage limits, Paid weekly Work and Codex rate limit resets, Using credits for flexible usage, ChatGPT release notes
• OpenAI: Codex pricing and rate card, Introducing ChatGPT Plus (1 February 2023), Introducing ChatGPT Pro (5 December 2024)
• Thibault Sottiaux (Tibo) on X, 29 October 2026: post
• Sam Altman on X, 5 January 2025 (US time): post
• Anthropic: Claude plans and pricing, What is the Ultra plan?, Higher usage limits and a SpaceX compute deal (6 May 2026), weekly limits announcement on X (28 July 2025)
• Cursor: Clarifying our pricing (4 July 2025)
• Reporting: BleepingComputer (1 November 2026, OpenAI email), The Decoder (1 November 2026)

Researched and drafted with AI assistance, including Anthropic’s Claude.