Elon Musk said on November 1, 2026 that Grok Bot's serving cost was 'unsustainably high relative to revenue' after subscribers found their weekly allowance used up. Here is how six other always-on and agentic AI tools handle limits and price.
After Grok Bot subscribers found their weekly allowance exhausted over Halloween weekend, Elon Musk said on November 1, 2026 that the cost of serving the agent was unsustainably high relative to revenue. Of the alternatives, Claude Cowork has the clearest published plan prices, while self-hosted Hermes Agent and OpenClaw are the only options with no vendor-set usage cap.
Key takeaways
Grok Bot's weekly usage pool is unpublished: Elon Musk confirmed on November 1, 2026 that cuts were a cost decision and not a glitch, and neither SpaceXAI nor reporters have given pool sizes.
Every hosted always-on agent compared here meters usage: Claude Cowork draws on the shared plan allowance, ChatGPT Dots is exempt from allowances only for its first month, and Perplexity Computer uses credits.
Claude Cowork is the only option here whose plan prices are on the vendor's product page: Pro about $17 a month billed annually ($20 monthly), Max $100, Ultra $200.
Manus sells its persistent Cloud Computer as a separate monthly plan, $30 (Standard) or $50 (Advanced), billed on top of the agent plan (per third-party trackers). It deletes the machine's files if you stop paying.
Hermes Agent and OpenClaw are MIT-licensed and free, so the cap is whatever your own model provider and server impose, but you run and secure them yourself.
Nov 1, 2026
Date Elon Musk replied on X that Grok Bot's serving cost was unsustainably high relative to revenue (TechCrunch, Teslarati)
Aug 11, 2026
Grok Bot beta launch date (MacRumors, third-party comparisons)
Sep 29, 2026
ChatGPT Dots launch date at OpenAI DevDay (SiliconANGLE)
$30 / $50
Monthly price of Manus Cloud Computer Standard and Advanced plans (Manus help centre)
Grok Bot is not the only always-on AI agent whose usage limits can change under you. Elon Musk said on November 1, 2026 that serving it was “unsustainably high relative to revenue”, and the hosted alternatives each cap usage in their own way. Claude Cowork has the clearest published pricing among the hosted rivals. The only options without a vendor-set cap are free, self-hosted agents that you run yourself.
TechCrunch reported on November 2 that some Grok Bot subscribers found their weekly allowance used up over Halloween weekend. One showed usage jumping from as low as 50% to 100% without overnight use. Musk’s reply on X, Teslarati’s same-day coverage and the absence of published pool sizes are the facts we can stand behind. How the limits change for each subscription tier has not been stated.
Limits are the product, not a footnote
Always-on agents run for hours on a cloud machine, so their cost per user is far above a chat reply. That is why every hosted product here meters usage. Anthropic says on its product page that Cowork “consumes limits faster than Chat”. TechCrunch reported on November 2 that after OpenAI merges ChatGPT’s Chat and Work tabs, ChatGPT and Codex will draw from the same allowance pool, with no rollout date given. ChatGPT Dots is in an introductory window: agent usage does not count against allowances for the first month, with later terms still to come.
What to check before you switch
Check three things. Is the allowance size published? What happens when it runs out: a wait, a metered charge or a hard stop? Does your work need a standing machine at all? A scheduled report or inbox sweep fits Claude Cowork’s task model. A site or bot that must stay up fits Manus’s flat-rate virtual machine or a self-hosted agent on a small server.
The self-hosted trade-off
Hermes Agent and OpenClaw remove the vendor’s weekly cap, not the cost. You still pay a model provider per use, and you are responsible for hosting, updates and what the agent is allowed to touch. A developer comfortable with that trades a surprise limit cut for a predictable bill. Everyone else is better served by a published price than by a promise of unlimited use.
Where this is heading
Grok Bot’s cut follows a run of launches within months: Grok Bot in August, Cowork’s merge into Claude on September 16, Dots on September 29. Prices for the standing-agent tier are still being set, and OpenAI says Dots terms and extra-agent pricing are to come. Expect further changes, and check each vendor’s page before committing to an annual plan.
Ranked tools
01
1. Claude Cowork
Best for published prices and scheduled tasks
Claude Cowork is included in Anthropic's paid plans. Anthropic's product page lists Pro at about $17 a month billed annually ($200 up front) or $20 monthly, Max at $100, Ultra at $200 with more Cowork usage, and Team at $20 per seat per month. Cowork can run tasks on a daily, weekly or monthly schedule and keeps running after you close your laptop. The desktop app is required only for access to folders and apps on your own computer. Cowork is task-based, not a standing agent with a long-lived machine: a third-party comparison dated October 19, 2026 says there is no single persistent agent. The page states the limit trade-off openly. Cowork "consumes limits faster than Chat" because agentic tasks coordinate sub-agents and tool calls, and heavier users are advised to upgrade.
Pros
Plan prices and the limit trade-off are published on the product page
Scheduled tasks keep running with the laptop closed
Works on web, mobile and desktop, and is bundled into plans you may already pay for
Cons
Draws on the same allowance as chat, so heavy agent use can use it up faster
Task-based, with no single standing agent identity
Anthropic notes the unified experience is still rolling out, so availability varies by account
02
2. ChatGPT Dots
Best for a standing agent inside ChatGPT, Slack or Teams
OpenAI launched Dots at DevDay on September 29, 2026, per SiliconANGLE. Each Pro or Business Premium plan includes one agent with its own cloud computer that you can open to watch, reach 4,000-plus apps through plugins, and message from ChatGPT, Slack or Teams. The catch for anyone worried about limits is that the terms are not final. Agent usage does not count against plan allowances for the first month, and what happens after that "will be listed later". Extra agents and capacity upgrades are planned at undisclosed prices. Pro users in the EEA, Switzerland and the UK are excluded. On November 2, 2026 TechCrunch also reported that OpenAI is merging ChatGPT's Chat and Work tabs, after which ChatGPT and Codex will draw from the same allowance pool. No start date was given.
Pros
Closest direct counterpart to Grok Bot, with a standing agent on its own cloud computer
Reachable from Slack and Microsoft Teams as well as ChatGPT
One agent is included in the plan, with an auto-review check on account-touching actions
Cons
Post-launch pricing and usage terms are not yet published
Pro plans in the EEA, Switzerland and the UK are excluded
Only one Dot per plan at launch
03
3. Manus Cloud Computer
Best for a dedicated, persistent machine at a fixed price
Manus sells a persistent Ubuntu Server 24.04 virtual machine separately from its agent subscription. Its help centre lists two monthly plans: Standard at $30 (2 vCPUs, 4 GB RAM, 70 GB storage) and Advanced at $50 (2 vCPUs, 8 GB RAM, 120 GB storage), each with a dedicated public IP. The price is a flat rate for the machine, but according to third-party trackers the Cloud Computer is billed in addition to the Manus agent subscription. The machine exists only while you pay: if the subscription lapses, the VM shuts down and working files, installed tools and databases are deleted. An older $10 Basic tier appears in third-party trackers but not on the current help-centre page, so check the live page.
Pros
Flat monthly price for the machine, with specs published
Dedicated public IP and full Ubuntu environment
Suits hosting sites, bots and scripts, not just chat tasks
Cons
Billed separately from the agent plan, so total cost is higher than the headline
Everything on the machine is deleted if payment lapses
Agent usage is still metered by the Manus plan itself
04
4. Grok Bot
The agent in the news, with the least transparent limits
Grok Bot is SpaceXAI's always-on agent, launched in beta on August 11, 2026. Several Bots share one cloud computer with a browser, files and a terminal. They can be taught routines and connect through MCP, connectors and plugins. Access comes with eligible Cursor plans or a linked SuperGrok or X Premium+ subscription. DataCamp describes only a seven-day trial credit, not a permanent free tier. Each plan draws on a weekly allowance that SpaceXAI has not published. On November 1, 2026 Musk replied on X that the cost of serving Grok Bot "was unsustainably high relative to revenue". TechCrunch reported that users had seen weekly usage jump to 100% without using the agent overnight. Musk has not said how limits will change for subscribers. Reports differ on which subscription tiers carry access and at what price, so confirm on your own account before paying.
Pros
Several Bots with distinct jobs sharing one machine, plus group chats
Connectors, MCP and routines for repeatable work
Included with paid Cursor plans, so existing subscribers may already have it
Cons
Weekly pool sizes unpublished, and limits were just cut
Cannot choose the underlying model, per a third-party comparison updated October 19, 2026
Reports conflict on tiers and pricing
05
5. Hermes Agent (Nous Research)
Best free, self-hosted agent with built-in scheduling
Hermes Agent is an MIT-licensed agent from Nous Research that builds skills from past tasks, keeps memory across sessions and has a built-in cron scheduler for unattended jobs. It runs on Linux, macOS, Windows and Android, connects to Telegram, Discord, Slack, WhatsApp, Signal and email, and supports local, Docker, SSH, Modal, Daytona and Vercel Sandbox backends. It works with many model providers, including OpenRouter and OpenAI, and the project says it can run on a $5 VPS. There is no vendor weekly cap, but you pay your model provider per use and you are responsible for setup and security.
Pros
Free and open source, with natural-language scheduled tasks
Many model providers, so you can switch models
Runs where you choose, from a small VPS to serverless
Cons
You manage hosting, updates and security
Model costs are metered by your provider, not capped for you
No vendor support or published reliability benchmarks
06
6. OpenClaw
Best for a local assistant you control from chat apps
OpenClaw is a self-hosted AI assistant that you message from chat apps such as WhatsApp or Telegram. Its site says it is MIT licensed with "no subscription, no hosted tier, no token", keeps state on your machine and remembers preferences and projects between conversations. It can use hosted, subscription-backed, gateway or local models and favours no lab's model in code. As with Hermes Agent, the usage limit is whatever your model provider charges or allows. A nonprofit foundation stewards it, which may matter if you worry about a vendor changing terms.
Pros
Free and MIT licensed, with no hosted tier to be repriced
State stays on your own machine
Bring your own model, including local ones
Cons
You operate and secure it yourself
Model usage costs are separate
Less suited to a team that wants a managed cloud computer
07
7. Perplexity Computer
Credit-based option for research-heavy tasks
Perplexity Computer runs multi-step tasks with scheduled recurring runs and meters them in credits. Perplexity's own page could not be read when we checked, so the plan details come from third-party trackers, which broadly agree that Max is $200 a month with a monthly credit allowance (reported as 10,000) but disagree on what Pro includes and on bonus credits. A Perplexity changelog dated March 13, 2026, cited by one of those trackers, says Computer reached all Pro subscribers. Treat the figures as provisional and confirm on Perplexity's pricing page.
Pros
Recurring Schedules for repeat tasks, per third-party guides
Available to Pro subscribers, not just the top tier
Credit metering makes the cost of a task visible
Cons
Credits can run out quickly on long tasks, according to third-party guides
Shared plan allowance; Cowork uses it faster than chat
Pro ~$17-20/mo, Max $100, Ultra $200
Anthropic cloud, desktop app for local files
ChatGPT Dots
Standing agent
Not counted for the first month; terms to be announced
Included in Pro and Business Premium; later pricing unpublished
OpenAI cloud computer
Manus Cloud Computer
Persistent VM
Agent plan is metered; the VM is flat-rate
$30 or $50 a month, plus the agent plan
Manus cloud, Ubuntu 24.04
Grok Bot
Several Bots on one machine
Unpublished weekly pool, recently cut
Eligible Cursor plans or linked SuperGrok / X Premium+; tiers reported inconsistently
SpaceXAI cloud computer
Hermes Agent
Self-hosted, scheduled
Set by your model provider
Free software; you pay model and server
Your machine or sandbox
OpenClaw
Self-hosted assistant
Set by your model provider
Free software; you pay model and hardware
Your machine
Perplexity Computer
Task-based, scheduled
Credits; figures vary by source
Max ~$200 per third-party trackers; Pro terms vary
Perplexity cloud
Frequently asked questions
Did Grok Bot's limits really get cut?
In effect, yes. TechCrunch reported on November 2, 2026 that subscribers found weekly usage maxed out. Elon Musk replied on X on November 1 that this was not a glitch because the cost of serving Grok Bot was unsustainably high relative to revenue. The weekly pool sizes have not been published.
What is Grok Bot?
Grok Bot is SpaceXAI's always-on AI agent platform, launched in beta on August 11, 2026. Several Bots share one cloud computer with a browser, files and a terminal, and keep working after the user logs off.
Which alternative has the clearest pricing?
Claude Cowork's product page lists Pro at about $17 a month billed annually or $20 monthly, Max at $100 and Ultra at $200, with Cowork using the plan's shared allowance. Manus also publishes its Cloud Computer prices, $30 and $50 a month, but that is separate from its agent plan.
Is there an always-on agent with no usage cap?
Not from a hosted vendor. Hermes Agent and OpenClaw are free, MIT-licensed and self-hosted, so the only limits are those of the model provider and server you choose, and you handle the setup and security yourself.
How does ChatGPT Dots compare with Grok Bot?
Both give an agent its own cloud computer. Dots, launched September 29, 2026, includes one agent in ChatGPT Pro and Business Premium and can be messaged in Slack or Teams, but OpenAI has not published post-launch usage terms. Grok Bot supports several Bots on one machine and has unpublished weekly limits.
How we evaluated
We started from the November 1-2, 2026 reports of Grok Bot's limit cuts, confirmed by TechCrunch and Teslarati, and then chose agents a reader could realistically switch to for scheduled or always-on work. We read vendor pages where they were available (Claude, OpenClaw, Hermes Agent, Manus help centre) and a launch report for ChatGPT Dots. Perplexity's page was not readable, so its details are flagged as third-party. Prices were checked on November 3, 2026 and change often.
Whether the vendor publishes plan prices and the size of the usage allowance
Whether the agent keeps working when you are offline and where it runs
What happens when the allowance runs out
Who has to set up and secure the agent
Innvesti may have commercial relationships with some companies mentioned. Our analysis is
editorially independent. Figures cited are sourced; where reliable data is unavailable we
say so rather than estimate.
