How Claude Cloud Session Credits Dey Work
Claude cloud sessions dey run Claude Code for Anthropic servers. See wetin credits cover, why chat, Projects and Routines no dey count, and when dem finish.
Why cloud session dey use credit when you already dey pay for plan
Cloud session credit na free bonus balance wey cloud session dey use first before e touch the usage wey your plan already include. Anthropic give am to eligible individual Pro and Max subscribers as promotion, so no be money wey you pay in. Cloud session na Claude Code session wey dey run for cloud instead of your laptop: by default, e dey run for machines wey Anthropic manage, or for organization self-hosted environment when the organization route sessions go there. E still belong to your subscription, and Anthropic documentation clearly talk say no separate charge dey for the virtual machine (VM) wey the session dey run on. Wetin the credit change na the order wey the meter dey read: while credit balance dey attached to your account, cloud sessions go use that balance first. Na only after e finish dem go start count against the same pool wey your terminal sessions and chats already dey share. The promotion credit expire on 4 November 2026, so cloud sessions now dey count against that shared pool from the beginning.
So the answer to “why this dey charge credit when I already dey pay for Pro, Max, or Ultra” na say the credit na extra pool wey dey before your plan, no be replacement for am. Nothing change about your subscription. Another meter just appear ahead of am.
This session dey run for your machine or for cloud?
Everything else depend on this difference, and interface no always make am clear. Session dey run for cloud when you start am from claude.ai/code for browser, from Code tab for Claude mobile app, from Desktop app with Cloud selected instead of Local, from terminal with claude --cloud, or from scheduled routine. Session dey run for your own machine when you start am for your terminal, for your IDE, or for Desktop app with Local selected.
One case look like exception, but e no be exception. Remote Control let you watch and control session from your phone or browser while that session dey execute for your own machine. Na local session, so e never use cloud credit, because the work dey run for hardware wey you already own.
Cloud sessions dey available for Pro, Max, Ultra, and Team plans, and for Enterprise users wey get premium seats or Chat + Claude Code seats. If custom roles dey control your access, you still need role wey get cloud sessions permission before you fit start sessions for Anthropic's infrastructure.
Cloud session usage dey come from where?
E dey come from the normal pool for your account, after any credit wey dey before am don finish. The cloud sessions documentation talk am clear: "cloud sessions share rate limits with all other Claude and Claude Code usage within your account. Running multiple tasks in parallel consumes more rate limits proportionately." The same page add say "there's no separate charge for the cloud VM."
Read the second sentence two times, because na that one dey surprise people. Each claude --cloud command dey create im own session with im own context window. If you start four tasks at once, four sessions go send their own requests at the same time, so the amount wey dem draw from your account go roughly be four times wetin one session go use. The VM free. The tokens no be free.
The exceptions na features wey bill usage credits instead of your plan. We cover dem for the next section.
The word "credit" mean two different things
The interface dey use one word for two separate mechanisms, and na this overloading dey cause most of the confusion behind this question.
Usage credits na the permanent mechanism. Na prepaid balance wey you fund yourself for Settings > Usage, and e dey available for Pro, Max, and Ultra. For most work, dem no do anything until you reach one of your plan limits: Anthropic describe dem as starting "instead of being blocked when you reach one of your plan's limits". Some model and feature choices na exception, and dem dey use am from the beginning, as the next paragraph explain. After that point, if credits dey on and you get funds, your work go continue and usage go be "billed at standard API pricing rates". API mean application programming interface, and standard API rates dey charge per token, so na metered spending, no be allowance. You fit set monthly spend cap, and /usage dey show the month's spending against am. The cap no be the only stop: Anthropic talk say usage go continue "until you reach your monthly spend limit, or your balance runs out with auto-reload off", so balance wey you fund once fit finish before you reach the cap. Usage credits cover both Claude conversations and Claude Code. For the full picture of this balance, including the second unrelated meaning wey the word carry, read wetin usage credits be and where the word dey get mixed meaning.
Some features dey spend usage credits even when plan usage remain. For Pro, Max, Ultra, Team, and Enterprise, fast mode dey "available via usage credits only and not included in the subscription rate limits", and the fast mode docs add say e "draws directly from usage credits, even if you have remaining usage on your plan". Until credits dey on, /fast go refuse with "Fast mode requires usage credits". Fast mode dey work for cloud sessions too. Ultrareview, wey dey run as cloud session, dey give Pro, Max, and Ultra three free runs per account, then each review go bill to usage credits. Model choice matter too. For Pro, Anthropic's Fable article talk say Fable models normally dey run on pay-as-you-go usage credits from the start, because dem no dey part of the plan usage limits. A Fable 5.2 launch offer change this for one week: from November 9 reach November 16, 2026 (11:59 pm PT), Pro fit "use up to 50% of your weekly usage limits on Fable models at no extra cost", and the offer cover "every Fable model, including Fable 5.2, Fable 5.1, and Fable 5." Anthropic add: "After you use that 50%, or once the week ends, Fable usage is billed to usage credits if you have them turned on." For Max and Ultra, Fable fit use up to half of your weekly limits at no extra cost, then e go continue with usage credits. Na their normal plan allowance be that. The launch offer included week apply only to Pro and standard Team seats. Eligible Pro and standard Team seats also get one Fable reset, wey expire November 16, 2026 at 11:59 pm PT. Max, Ultra, premium Team seats, and premium seats on legacy seat-based Enterprise plans no get included launch week; eligible accounts receive one to three one-time Fable resets, depending on subscription length, wey expire March 8, 2027 at 11:59 pm PT. Na these cloud work dey cost money on top of your plan.
A cloud session credit na promotional bonus, no be prepaid balance. Anthropic's help center describe am as bonus credit of $100 for Pro and $250 for Max, and e talk say "Cloud sessions are included in your Pro or Max plan, and this credit adds free usage on top of your regular plan limits." Dem grant am, dem no sell am, and e cover only cloud sessions wey you start yourself: Projects and Routines no dey included even though routines dey run as cloud sessions, and chat, Cowork, and Remote Control sessions no dey included too. The system apply am automatically when cloud session start. While e last, cloud session usage dey paid from am and e no count toward your plan limits. One special matter for Pro: the terms talk say "Your bonus credit is used first, but Fable will draw on your usage credits immediately after the bonus credit is spent." To select Fable for Pro, usage credits must dey turned on with balance, even while bonus credit still remain.
The monthly API credits wey now come with Max ($100 per month) and Ultra ($200 per month) separate from both balances above. Dem go to linked Claude Console organization, and dem no fit pay for interactive Claude Code or usage credits inside Claude, Claude Code, or Cowork.
For here, the honest answer be say dem document the two for different places, so check your own account instead of trusting any summary, including this one. As of 10 November 2026, the cloud sessions page for code.claude.com cover rate limit sharing and no talk anything about promotional credit, while the usage credits help article cover the prepaid balance and no talk anything about cloud sessions. The promotion get its own help center article. E open to individual Pro and Max subscribers wey their subscription dey active when e start on 23 September 2026; Team, Enterprise, free trial, and past-due accounts no qualify, and sign-ins through Amazon Bedrock, Google Vertex AI, or an API key no qualify too. The claim deadline na 7 October 2026, and the credit expire on 4 November 2026 at 11:59 PM PT. Until then, the remaining balance dey show for the usage menu.
Wetin go happen when pool don empty?
Three different things fit happen, depending on which pool empty.
When promotional cloud session credit finish, nothing stop. Cloud sessions go back to count against your normal plan limits, except Fable for Pro, wey go straight to usage credits because Fable normally dey outside Pro plan limits. For one week, that rule get exception: from November 9 through November 16, 2026, the Fable 5.2 launch offer include every Fable model (Fable 5.2, Fable 5.1, and Fable 5) inside Pro weekly limits, up to 50% of dem. So Pro Fable usage first dey use that included allowance, then e move go usage credits only after the 50% finish or the week end. One earlier promotion wey count Fable 5 toward Pro weekly limits end July 19, 2026, and e never apply to Fable 5.1 or 5.2. Na this case dey important to watch, because na the credit make cloud work feel free, and the switch back happen quietly.
When your plan included limit reach and usage credits dey on with money available, work continue and dem bill am per token at standard API rates until you reach your monthly spend limit, or until your balance finish if auto-reload dey off. Auto-reload, wey buy more credit when balance fall below threshold wey you set, close the second gap. When you reach spend limit, Claude Code go show message wey name the limit wey e hit, like "You've hit your monthly spend limit". For Pro, Max, and Ultra, you fit raise the cap for Settings > Usage, wey /usage-credits open for your browser.
When your plan limit reach and usage credits dey off or no get funds, new work stop. For eligible terminal, IDE, and Desktop Code sessions wey dey run Claude Code 2.1.277 or later, small allowance wey get fixed cap dey roll out. E allow response wey don already start to reach stopping point after five-hour limit reach; e still count toward weekly limit. Verified Ultra accounts no get five-hour limit; accounts wey never verify get temporary limit. You go see "You've hit your session limit" or "You've hit your weekly limit", together with the time wey that window go reset. Those limits cover every model, so /model no go save you. Model-specific message dey behave differently: after "You've hit your Opus limit", switching to another family with /model sonnet go let you continue work. When that limit na the weekly Opus allowance, wetin e mean when Opus priority access finish explain when Opus go come back and whether credits or Max upgrade fit restore am earlier. The options wey dey available when limit land and how dem dey count reset windows both easier to read before you need dem than during the problem.
Nothing for this path dey quietly downgrade your model or reduce your context just to continue. The meter either get something wey e fit draw on, or e stop. If the message wey stop you mention Claude's safeguards instead of a limit, credits and reset times no be the issue. Wetin safeguards message mean for the session and your account cover the next step wey Anthropic name.
Where you fit see wetin cloud session consume?
Two places dey, but one of dem fit mislead you if you read am wrong.
Settings > Usage for claude.ai na the authoritative view. E show your plan usage, whether usage credits dey on, your balance, this month's spend, and your monthly spend limit. From terminal, /usage-credits go open that same page for your browser.
/usage inside a session dey show your plan usage bars. For Pro, Max, Ultra, Team, and Enterprise, e also add breakdown of where recent usage go: skills, subagents, plugins, and individual MCP servers, each as percentage of the total. E also flag behaviours wey account for large part of recent usage, like long context or cache misses. Press d or w to switch between the last day and the last week.
Now na the trap. E dey calculate those day and week figures from session history for the machine wey you type the command on. Documentation talk am directly: "usage from other devices or claude.ai is not included." So if you run /usage for your terminal after heavy cloud work for afternoon, e go show quiet day, because the cloud work happen for another place. Open Settings > Usage when you want the cloud figures.
Wetin dey make cloud session cost plenty?
The cost of any Claude Code session na the size of the context wey e send multiplied by how many times e send am. Cloud sessions dey increase both sides, for reasons wey no relate to the VM.
Long background run dey resend the whole conversation for every turn, and each tool call dey add another request wey carry that batch of results. So, session wey dey work for one hour dey pay for its own history many times. Na the same mechanism wey wetin really dey consume tokens for Claude Code session describe, and e apply the same way to local work. The difference na say nobody dey watch cloud session, so e dey run much longer before person stop am.
Re-reading the repository na another major cost. The VM dey clone the repository fresh and e no get memory of wetin you explore yesterday. Vague instruction like "clean up the error handling" go make Claude search wide and read plenty files just to find where the error handling dey. Specific instruction like "add the retry wrapper from src/http/retry.ts to the three callers in src/sync/" dey send only small part of the tokens, because e remove the search work.
/clear dey work for cloud session outside project threads the same way e dey work locally: e starts fresh conversation with empty context for the same cloud session. Clearing between unrelated tasks dey keep the next task context small. Thread wey you continue to reuse without clearing dey carry everything wey happen before into every later request.
Idle gaps dey cost money too. Cloud session dey stop after e remain inactive for some time, and dem dey reclaim its VM. When you reopen the session from claude.ai/code, e provisions fresh VM and restores your conversation history, but background work wey still dey run when the VM disappear no dey restore. Your first message after that gap also no dey use the prompt cache, so the full context dey process again at full price instead of reading am back cheaply. By default, cache for the main conversation lasts one hour while you dey inside your plan, and five minutes once you dey use usage credits. But that na default, no be fixed rule: the promptCacheTtl setting or the CLAUDE_CODE_PROMPT_CACHE_TTL environment variable set to 1h keeps the one-hour duration under either billing mode, but cache writes go cost higher rate. For cloud session, environment variables dey go on the cloud environment. Subagents dey default to five minutes under both modes.
Controls wey dey work inside cloud session outside project threads include: /context to see wetin dey inside the window, /clear to start again with empty context, /compact keep the test output to summarize with focus wey you choose, and /model sonnet or /effort with a level as argument, so you no go continue to pay Opus rates for mechanical work. Note say /compact dey read the conversation wey e summarize, so compacting very large context na itself a large request. Subagents dey keep verbose output for separate context window, but each subagent dey send its own requests. So dem dey reduce your main context, not your total context. Wetin agent harness dey add to every request cover the shape of that overhead.
Run the same job locally
The local way to do this work na the ordinary CLI, and the useful pattern na to split the job instead of choosing only one side.
Plan for your machine, where you fit stop exploration easily:
claude --permission-mode planFor plan mode, Claude dey read files and run commands to explore, then e propose an approach without editing source. Commit the plan, push am, then hand the execution over to a cloud session:
claude --cloud "Execute the migration plan in docs/migration-plan.md"One detail dey catch everybody at least once. The cloud VM clones your repository's GitHub remote at your current branch, not your local working copy, so anything wey you never push no dey exist as far as that session concern. Push first. The exception na repository wey get no git remote, or a github.com repository wey Claude GitHub App never install for: for that case Claude Code uploads a bundle of your local repository instead. The bundle include uncommitted changes to tracked files, but e no include untracked files. For macOS, Linux, and WSL, the bundle leaves out uncommitted changes to files wey get names like credentials or keys, including .env, *.tfvars, and *.pem, and files wey git filters manage, like Git LFS. Native Windows uploads tracked changes as dem be.
To finish a cloud session for your own machine, pull am down:
claude --teleportTeleport fetches and checks out the session branch, then e loads the conversation into your terminal. If the cloud session was compacted, e continues from the summary instead of copying the earlier messages. E requires clean working tree and checkout of the same repository, not a fork. After that, the terminal get its own copy of the session. New work for there stays local and no go show back for claude.ai.
Running locally no make any single token cheaper. Wetin e gives you na control: /resume to return later, no VM expiry, and the ability to press Escape immediately Claude start going the wrong way. For work wey you go sit near anyway, that control usually worth more than the convenience. For test suite wey dey take twenty minutes, or task wey you want make e continue while your laptop dey shut down, cloud get good reason to dey there. Keeping the context small helps for both places, and na that deliberate context window management be the topic.
When changing plan na the real answer
Credits na the expensive way to buy capacity, because dem dey charge per token with standard rates while subscription na flat fee. If you dey top up every week, the calculation don already turn against you.
Two signs show say na the plan be the problem, not your habits. The first one na when you dey reach the limit during normal work, before you try anything unusual. The second one na credit balance wey dey finish on the same schedule every month. More precise prompts no go fix either problem.
Wetin to compare, and where the plans really differ, belong for the plan pages instead of one paragraph here. Start with which plan match the way you dey work, and if you dey use Max already, the difference between Max and Ultra. Make your decision with your own figures from Settings > Usage open in front of you.
FAQ
Cloud sessions dey cost extra on top my Claude plan?
No. The cloud sessions documentation talk say “there's no separate charge for the cloud VM” and cloud sessions “share rate limits with all other Claude and Claude Code usage within your account”. Dem include the machine. Wetin you dey spend na tokens, wey come from the same pool wey your terminal and chats dey use, unless bonus credit dey before that pool and dem spend am first. Exceptions na work wey usage credits dey bill, like fast mode, ultrareview runs after the free ones, and Fable for Pro after its launch allowance. Dem cost extra anywhere dem run. Until November 16, 2026, Pro include all Fable models up to 50% of weekly limits without extra cost.
Why my local /usage no show wetin my cloud session do?
Because e dey calculate the day and week figures from session history for the machine wey you run the command on. The documentation clear say “usage from other devices or claude.ai is not included”, so cloud work no go show there. Open Settings > Usage for claude.ai. Na there be the authoritative view for both plan usage and credit spend.
My cloud session go stop when the credit finish?
E depend on which credit. When the promotional cloud session credit finish, or expire on 4 November 2026, cloud sessions continue to run and return to counting against your normal plan limits. Fable for Pro normally dey bill to usage credits, but from November 9 through November 16, 2026, the Fable 5.2 launch offer include all Fable models for Pro up to 50% of weekly limits. Usage credits apply after you use that allowance or the offer end. When you reach your plan limit and usage credits dey enabled and funded, work continue. Dem bill am per token at standard API rates until your monthly spend limit, or until your balance finish if auto-reload dey off. When you reach your plan limit and no credits dey available, new work stop with message like “You've hit your session limit” wey name the reset time. Anthropic document the capped five-hour wrap-up allowance for eligible terminal, IDE, and Desktop Code sessions, no be cloud sessions.
Claude Code cheaper to run locally than for cloud?
Per token, no. The same request cost the same wherever the session execute, and the VM free either way. For real use, local sessions often use less because you dey watch dem, you fit stop a run wey dey go wrong, and no VM expiry force restart wey go process your history again. Cloud session fit cost more when you leave am to work unattended with vague instruction.