[{"data":1,"prerenderedAt":63},["ShallowReactive",2],{"$f3v1lkwlcbyvrf":3},{"href":4,"title":5,"description":6,"kind":7,"mark":7,"planned":8,"contributors":9,"provenance":7,"express":8,"bridge":8,"html":10,"headings":11},"\u002Fdocs\u002Fcontrol\u002Fspend","Spend and models","What counts as spend, the caps that hold it, which model a run uses, and the model connections that pay for it.",null,false,[],"\u003Ch2 id=\"your-keys-your-bill\">Your keys, your bill\u003C\u002Fh2>\n\u003Cp>Runs use model connections you add: your own Claude, OpenAI, Grok or OpenRouter credentials. The model provider bills you\ndirectly. Zero Human does not resell model usage, and a cheap model is not a budget: caps are.\u003C\u002Fp>\n\u003Ch2 id=\"what-counts-as-spend\">What counts as spend\u003C\u002Fh2>\n\u003Cp>Spend is the cost of model calls: every step of a run, and every chat reply. Tool calls are not spend. Each call is\npriced once, and the same figure is used by every cap, by the run's \u003Cstrong>Cost\u003C\u002Fstrong>, by today's total on \u003Cstrong>Spend\u003C\u002Fstrong>, and by\nan execution's total.\u003C\u002Fp>\n\u003Cdiv class=\"prose__table\">\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Connection\u003C\u002Fth>\n\u003Cth>What a call costs\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>OpenRouter API key\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>What OpenRouter reports it billed, plus the provider's own charge when you bring that provider's key to OpenRouter.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Claude Console API key\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>The model's list price for the tokens used, cache reads and writes included.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Claude Pro\u002FMax seat\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>A share of the seat's price: the seat's weekly price, times the share of its weekly limit the calls used, spread across them by what they would have cost on an API key. A seat never costs more than an API key would. Usage past the seat's limit is priced at API rates.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>OpenAI API key\u003C\u002Fstrong>, \u003Cstrong>xAI API key\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>The model's list price for the tokens used, cached prompt tokens at the cache price. A model with no published price is recorded at no cost, and its cost source says \u003Ccode>unpriced\u003C\u002Fcode>.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>ChatGPT seat\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>A share of the seat's price, worked out as for a Claude seat, from the weekly limit ChatGPT reports with each reply.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>SuperGrok seat\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>A share of the seat's price. Grok does not report how much of the plan is used, so each call is charged the most it could have cost: never more than an API key would have, and less the more the seat runs.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\n\u003C\u002Ftable>\n\u003C\u002Fdiv>\n\u003Cp>The seat's plan comes from the provider where it says (Anthropic, and ChatGPT when you sign in), or you set it on the\nconnection (\u003Cstrong>Seat plan\u003C\u002Fstrong> and \u003Cstrong>Monthly price (USD)\u003C\u002Fstrong>). A SuperGrok plan is always set by hand. Until a seat has a\nprice, its calls are costed at API list price.\u003C\u002Fp>\n\u003Cp>A run's \u003Cstrong>Cost\u003C\u002Fstrong> is at the top right of its page. Hover it for the split between API and seat usage and the tokens\nin, out and cached.\u003C\u002Fp>\n\u003Ch2 id=\"caps\">Caps\u003C\u002Fh2>\n\u003Cp>\u003Cstrong>Settings → Spend\u003C\u002Fstrong> holds your caps. A cap sits on one layer: the enterprise, a team, a member, a role, or a task.\nAmounts are in US dollars.\u003C\u002Fp>\n\u003Cdiv class=\"prose__table\">\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Limit\u003C\u002Fth>\n\u003Cth>What it holds\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Day $\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>Money spent today (UTC).\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Runs \u002F day\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>Runs started today.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Run $\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>What one run should cost. See \u003Ca href=\"#per-run-budgets\">Per-run budgets\u003C\u002Fa>.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Max model tier\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>The most capable \u003Ca href=\"#model-tiers\">tier\u003C\u002Fa> a run may use.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\n\u003C\u002Ftable>\n\u003C\u002Fdiv>\n\u003Cp>A run is held by every cap that applies to it: the enterprise's, and those on its teams (every team its member sits\non or leads), its member, its task and its member's roles. The tightest one wins.\u003C\u002Fp>\n\u003Cp>Today's spend and today's runs are counted for the \u003Cstrong>whole enterprise\u003C\u002Fstrong>. A day cap on a team or a task is compared\nwith everything the enterprise spent today, not only that team's or task's share; the Spend page says so on each row.\u003C\u002Fp>\n\u003Cp>Your enterprise has no caps until you add one. A new team or member copies the enterprise's \u003Cstrong>Day $\u003C\u002Fstrong> onto a cap\nof its own when it is created, with the enterprise cap's \u003Cstrong>Max model tier\u003C\u002Fstrong> and no other limit; while the enterprise\nhas no \u003Cstrong>Day $\u003C\u002Fstrong>, it gets none. It is a copy, made at creation: later changes to the enterprise cap do not flow into\nit, and you can change or clear it like any other cap.\u003C\u002Fp>\n\u003Cp>A team adopted from the catalog gets the daily cap you choose when you adopt it, and so does a teammate created when\nyou adopt a role. On the \u003Cstrong>Catalog\u003C\u002Fstrong>, the clone form's daily spend cap starts at the enterprise's \u003Cstrong>Day $\u003C\u002Fstrong> ($10 while\nthe enterprise has no \u003Cstrong>Day $\u003C\u002Fstrong> or it cannot be read, and never above the form's $10,000 limit), and you can change it\nbefore you clone: the cap saved is the one the form shows. With none chosen (over the API), each copies the\nenterprise's \u003Cstrong>Day $\u003C\u002Fstrong> the same way, or gets $10 a day while the enterprise has no \u003Cstrong>Day $\u003C\u002Fstrong>. Whenever the enterprise\nhas a cap, the adopted team's or teammate's cap carries the enterprise cap's \u003Cstrong>Max model tier\u003C\u002Fstrong>, whichever \u003Cstrong>Day $\u003C\u002Fstrong> it\nstarts with. The teammates a team adoption creates copy the enterprise's \u003Cstrong>Day $\u003C\u002Fstrong> like any new member.\u003C\u002Fp>\n\u003Ch3 id=\"when-a-cap-is-reached\">When a cap is reached\u003C\u002Fh3>\n\u003Cp>The caps are checked when a run is about to start, before its first model call. If today's spend (with room for the\nnext call) is at a day cap, or today's runs are past a \u003Cstrong>Runs \u002F day\u003C\u002Fstrong> cap, the run does not start. It is parked as\n\u003Ccode>paused_spend\u003C\u002Fcode>:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>It keeps its place: its task and its member stay busy, so later starts of that task wait or are skipped.\u003C\u002Fli>\n\u003Cli>It appears on \u003Cstrong>Blockers\u003C\u002Fstrong> under \u003Cstrong>Paused spend\u003C\u002Fstrong>, with \u003Cstrong>Raise cap\u003C\u002Fstrong>.\u003C\u002Fli>\n\u003Cli>A run already under way is not stopped by a day cap. It carries on to the end.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Paused runs start again when you raise a \u003Cstrong>Day $\u003C\u002Fstrong> cap to a higher amount (every paused run in the enterprise is\nqueued again and checks the caps afresh) or turn spend limits \u003Cstrong>Off\u003C\u002Fstrong>. Changing \u003Cstrong>Runs \u002F day\u003C\u002Fstrong> or \u003Cstrong>Run $\u003C\u002Fstrong> does not\nqueue them again, and nothing restarts them at midnight: raise a day cap, turn limits off, or cancel them. Once spend reaches 80% of the tightest day cap, \u003Cstrong>Spend\u003C\u002Fstrong> and \u003Cstrong>Enterprise\u003C\u002Fstrong> warn you and \u003Cstrong>Spend\u003C\u002Fstrong> in the\nsidebar shows a badge.\u003C\u002Fp>\n\u003Cp>Chat replies count toward today's spend too. Before each reply, the day caps on the enterprise and on that member\nare checked; at a cap the member says so and does not reply. See \u003Ca href=\"\u002Fdocs\u002Fwork\u002Fchat\">Chat and plans\u003C\u002Fa>.\u003C\u002Fp>\n\u003Ch3 id=\"per-run-budgets\">Per-run budgets\u003C\u002Fh3>\n\u003Cp>A task's \u003Cstrong>Spend USD \u002F run\u003C\u002Fstrong>, and \u003Cstrong>Run $\u003C\u002Fstrong> on a cap, set what one run should cost at most; the tighter of the two\napplies. It is only partly enforced today. On an OpenRouter connection, each model reply is kept short enough for the\nrun to stay within it, and a reply that the limit cuts off fails the run (\u003Ccode>llm_max_tokens\u003C\u002Fcode>). On a Claude connection\nit is not enforced during a run.\u003C\u002Fp>\n\u003Ch3 id=\"turning-spend-limits-off\">Turning spend limits off\u003C\u002Fh3>\n\u003Cp>The switch at the top of \u003Cstrong>Spend\u003C\u002Fstrong> turns spend limits \u003Cstrong>On\u003C\u002Fstrong> or \u003Cstrong>Off\u003C\u002Fstrong> for your enterprise. Off ignores every money\ncap, every run-count cap, the caps' model tiers and the tasks' per-run budgets, for runs and chat alike, and queues\nany paused runs again. Spend is still recorded, and a task's own highest tier still applies. With limits off, the\nrest of the Spend page and the \u003Cstrong>Paused spend\u003C\u002Fstrong> section on Blockers are hidden.\u003C\u002Fp>\n\u003Cdiv class=\"prose__planned\">\n\u003Cp class=\"prose__flag\">Planned\u003C\u002Fp>\n\u003Ch2 id=\"caps-still-to-come\">Caps still to come\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Cstrong>A month cap.\u003C\u002Fstrong> \u003Cstrong>Month $\u003C\u002Fstrong> can be saved on a cap today, but nothing holds a run to it yet.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Stopping a run mid-way.\u003C\u002Fstrong> A run that reaches a cap will stop before its next paid model call, instead of only\nnew runs being held.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>A per-run cap on every connection.\u003C\u002Fstrong> A run that reaches its per-run budget will stop, on a Claude connection as\non OpenRouter.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>An approval threshold.\u003C\u002Fstrong> A run that would spend more than an amount you set will wait at a gate for you before\nits next paid call.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003C\u002Fdiv>\n\u003Ch2 id=\"model-tiers\">Model tiers\u003C\u002Fh2>\n\u003Cp>A tier names a class of model, so the OS can choose &quot;a cheap one&quot; or &quot;the best one&quot; without naming it:\u003C\u002Fp>\n\u003Cdiv class=\"prose__table\">\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Tier\u003C\u002Fth>\n\u003Cth>For\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\n\u003Ctr>\n\u003Ctd>\u003Ccode>economical\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>The cheapest model; fine for API and data work.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode>mid\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>A good mid-priced model.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode>frontier\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>The most capable, and the most expensive.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\n\u003C\u002Ftable>\n\u003C\u002Fdiv>\n\u003Cp>Each tier stands for one model on each provider. The cap form shows which model a tier means on your connection.\u003C\u002Fp>\n\u003Ch3 id=\"which-model-a-run-uses\">Which model a run uses\u003C\u002Fh3>\n\u003Cp>A task's \u003Cstrong>Model\u003C\u002Fstrong> field decides:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>Auto\u003C\u002Fstrong>, which every new task starts on, has the OS choose. As the run starts, a small, fast model reads what the\nrun was asked, the task's instructions and its tools, and picks the tier the work needs: \u003Ccode>economical\u003C\u002Fcode> for simple,\nwell-specified work, \u003Ccode>mid\u003C\u002Fcode> for everyday work, \u003Ccode>frontier\u003C\u002Fcode> for the hardest. The run uses that tier's model on your\nconnection, and its page says which tier was chosen.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>An exact model\u003C\u002Fstrong>, picked from your providers' directories, is the model every run asks for. A tier cap does not\nblock it; money caps still apply. When the connection that answers does not offer that model, the run uses the\n\u003Ca href=\"#when-a-connection-does-not-offer-the-model\">closest model\u003C\u002Fa> it does offer.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Auto never chooses above the tightest \u003Cstrong>Max model tier\u003C\u002Fstrong> that applies to the run: the task's own, and the one on\nevery cap that covers it. With none set, it may choose any tier. Every cap carries a Max model tier, \u003Ccode>mid\u003C\u002Fcode> unless\nyou choose otherwise, so adding a cap also sets how high Auto may go.\u003C\u002Fp>\n\u003Cp>What else to know about Auto:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>Who chooses.\u003C\u002Fstrong> The choice is made on your own connection: by Jev on OpenRouter, by the Decisions API on an OpenAI\nAPI key, and by the provider's smallest model on Claude, Grok and a ChatGPT plan. The call is charged to the run. It\nreads a few thousand tokens and writes almost none.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>When nothing answers.\u003C\u002Fstrong> If the choice cannot be made within a few seconds, the run uses \u003Ccode>mid\u003C\u002Fcode>, or the highest\ntier it is allowed when that is lower. Its page says so.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>When there is nothing to choose.\u003C\u002Fstrong> With one tier allowed, or one model behind every allowed tier, nothing is\nasked.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>A run that carries on.\u003C\u002Fstrong> A run resumed from an earlier attempt stays on the model that attempt used.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>A task that drives a browser\u003C\u002Fstrong> runs on the highest tier it is allowed, with nothing chosen: a cheaper tier's\nmodel may not be able to operate one.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3 id=\"which-model-a-chat-uses\">Which model a chat uses\u003C\u002Fh3>\n\u003Cp>Chat is always on Auto. Each message you send has its tier chosen the same way, from the message and the last few\nturns of the conversation, up to the Max model tier of the caps that apply.\u003C\u002Fp>\n\u003Cp>A conversation holds the tier it is on for an hour after each turn. Within that hour it can move up for a harder\nquestion but not down, since a step down would re-read the whole conversation on a model that has none of it\ncached. After an hour of quiet the next message is chosen afresh. When your day cap is nearly spent, a turn comes\ndown to a tier it still has room for.\u003C\u002Fp>\n\u003Ch2 id=\"model-connections\">Model connections\u003C\u002Fh2>\n\u003Cp>\u003Cstrong>Settings → Agents\u003C\u002Fstrong> lists every model connection in the enterprise, whose it is, and whether it can be read. Each\nowner's connections are listed by priority, lowest number first; at the same priority, in the order a run tries\nproviders. Keys are never shown, only their last four characters. \u003Cstrong>Add Agent\u003C\u002Fstrong> adds one:\u003C\u002Fp>\n\u003Cdiv class=\"prose__table\">\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Provider\u003C\u002Fth>\n\u003Cth>How you connect it\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Claude\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>A Console API key, a setup token, or \u003Cstrong>Connect Claude Pro\u002FMax\u003C\u002Fstrong> to sign in with your seat.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>OpenAI\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>An OpenAI API key, or \u003Cstrong>Connect ChatGPT plan\u003C\u002Fstrong> to sign in with your ChatGPT plan (the login Codex uses).\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Grok\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>An xAI API key, or \u003Cstrong>Connect SuperGrok plan\u003C\u002Fstrong> to sign in with your SuperGrok plan (the login Grok Build uses).\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>OpenRouter\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>An API key.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\n\u003C\u002Ftable>\n\u003C\u002Fdiv>\n\u003Cp>Signing in with a ChatGPT plan ends on a page that does not load. That is expected: OpenAI sends your browser to\nan address on your own computer, where nothing is listening. Copy the whole address from the address bar, paste it\ninto the form, and choose \u003Cstrong>Finish connect\u003C\u002Fstrong>.\u003C\u002Fp>\n\u003Cp>Signing in with a SuperGrok plan ends on xAI's own page, which shows a code under \u003Cstrong>Enter this code to finish\nsigning in\u003C\u002Fstrong> and says to paste it into Grok Build. Paste that code into the form instead, and choose \u003Cstrong>Finish\nconnect\u003C\u002Fstrong>. The page's address is not the code: pasting it is refused with a message saying so.\u003C\u002Fp>\n\u003Cp>Pressing \u003Cstrong>Connect\u003C\u002Fstrong> again while a sign-in is under way reopens the same sign-in page rather than starting another.\nA sign-in that has expired says so when its code is pasted, and Connect then starts a new one.\u003C\u002Fp>\n\u003Cp>A connection belongs to the \u003Cstrong>enterprise\u003C\u002Fstrong>, shared by everyone; to a \u003Cstrong>team\u003C\u002Fstrong>, shared by its current members and\nlead; or to one \u003Cstrong>member\u003C\u002Fstrong>, used for that member's runs and chats. Add team connections on the Team page or choose\nthe team as the owner on \u003Cstrong>Settings → Agents\u003C\u002Fstrong>. If a task specifies an exact model, the OS first looks for that model on an eligible connection with\ncapacity, using that connection's own spelling of the model id.\u003C\u002Fp>\n\u003Cp>If no eligible connection with capacity lists that model, or the task specifies a tier instead, the OS tries:\u003C\u002Fp>\n\u003Col>\n\u003Cli>The member's own connections, lowest priority number first across providers.\u003C\u002Fli>\n\u003Cli>The task's teams the member currently sits on or leads, lowest priority number first across providers.\u003C\u002Fli>\n\u003Cli>The enterprise's connections, lowest priority number first across providers.\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>A member chat can use the teams that member sits on or leads. A task uses only its own teams. Removing the member\nfrom a team removes access to that team's connections unless they still lead it. Closed teams are excluded.\u003C\u002Fp>\n\u003Cp>For example, Grok at priority \u003Cstrong>0\u003C\u002Fstrong> comes before Claude at \u003Cstrong>5\u003C\u002Fstrong> and OpenAI at \u003Cstrong>10\u003C\u002Fstrong> for the same owner.\nAt equal priorities, Claude, OpenAI, Grok, then OpenRouter breaks the tie. A connection whose usage cannot be\nread can still be tried; an unknown usage reading does not mean its quota is exhausted.\u003C\u002Fp>\n\u003Cp>Each provider has its own model for each tier. A task that names a tier runs that provider's model for it, and you\ncan change the model behind a tier per provider.\u003C\u002Fp>\n\u003Cp>If the connection that answers does not list a tier's model (the provider retired it, or that account was never given\nit), the tier runs the next best model that connection does list: the one most alike in name and nearest in output\nprice. The cap form shows the model each tier is running. Nothing is blocked, and the tier goes back to its own model\nas soon as the connection lists it.\u003C\u002Fp>\n\u003Cp>The OS's own tasks have no assigned member and borrow no team connections, so their fallback order is the enterprise's connections first,\nthen the canonical COO's connections. An exact model is still tried first when an eligible account offers it.\u003C\u002Fp>\n\u003Ch3 id=\"when-a-connection-does-not-offer-the-model\">When a connection does not offer the model\u003C\u002Fh3>\n\u003Cp>When no eligible connection with capacity offers the exact model, the first connection in the fallback order\nwith capacity answers the run, using the closest of the three models your \u003Ca href=\"#model-tiers\">tiers\u003C\u002Fa> mean on that\nprovider. Model ids are translated between providers: \u003Ccode>anthropic\u002Fclaude-sonnet-5.5\u003C\u002Fcode> on OpenRouter is\n\u003Ccode>claude-sonnet-5-5\u003C\u002Fcode> on Claude.\u003C\u002Fp>\n\u003Cp>The closest tier model is chosen as follows:\u003C\u002Fp>\n\u003Col>\n\u003Cli>The tier you gave the named model yourself, if it is one of your tier models on any provider.\u003C\u002Fli>\n\u003Cli>Otherwise the tier model whose output price is nearest to the named model's.\u003C\u002Fli>\n\u003Cli>If nothing prices the named model, the tier the run would have used with \u003Cstrong>Default\u003C\u002Fstrong>.\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>The run page shows the model the run used, and the model it stands in for. The exact-model picker includes the models offered by enterprise, team and individual connections. An exact\nmodel available only on OpenRouter is tried there before substituting another model on a higher-priority connection.\u003C\u002Fp>\n\u003Ch3 id=\"when-a-connection-cannot-be-used\">When a connection cannot be used\u003C\u002Fh3>\n\u003Cp>A run never fails for want of a model. It is \u003Cstrong>blocked\u003C\u002Fstrong>, says why, and resumes when the cause clears:\u003C\u002Fp>\n\u003Cdiv class=\"prose__table\">\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>What happened\u003C\u002Fth>\n\u003Cth>Reason\u003C\u002Fth>\n\u003Cth>Resumes\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\n\u003Ctr>\n\u003Ctd>No connection at all\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>llm_not_bound\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>As soon as a connection the run can use is added: the member's own or the enterprise's. It does not try again before then\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>The provider refused the credential, or it is out of credit\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>llm_http_401\u003C\u002Fcode>, \u003Ccode>…402\u003C\u002Fcode>, \u003Ccode>…403\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>As soon as you change a credential, and it tries again hourly. If the refused credential is a plan sign-in, the next row applies instead\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>A plan sign-in (Claude Pro\u002FMax, ChatGPT plan, SuperGrok plan) has expired or been withdrawn\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>claude_oauth_refresh_failed:…\u003C\u002Fcode>, \u003Ccode>llm_oauth_refresh_failed:…\u003C\u002Fcode>, or \u003Ccode>llm_http_401\u003C\u002Fcode>\u002F\u003Ccode>…403\u003C\u002Fcode> on that sign-in\u003C\u002Ftd>\n\u003Ctd>At once, on another member of the task's role whose sign-in works. If nobody's works, as soon as one of them signs in again. It does not try again hourly (see the note under this table)\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>A rate limit is used up: a Claude or ChatGPT seat's session or weekly window, or a Claude model window\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>llm_quota_exhausted\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>When the window resets\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>A spend limit is used up: Claude's extra usage, or an OpenRouter key limit that resets daily, weekly or monthly\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>llm_quota_exhausted … kind=spend_limit\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>When the limit resets: at the time the provider gives, or else the next midnight, Monday or 1st of the month (UTC)\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>A balance is used up: prepaid credits, or a key limit that never resets\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>llm_quota_exhausted kind=balance\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>It tries again hourly: add credits\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>The model does not exist\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>llm_http_404\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>When the model answers again; after an hour it waits for you\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\n\u003C\u002Ftable>\n\u003C\u002Fdiv>\n\u003Cp>\u003Cstrong>A sign-in that stopped working.\u003C\u002Fstrong> Only a person signing in again fixes a plan sign-in that has expired or been\nwithdrawn, so the OS does not keep trying it. \u003Cstrong>Blockers\u003C\u002Fstrong> lists the sign-in, naming whose it is, with a\n\u003Cstrong>Reconnect\u003C\u002Fstrong> action that opens \u003Cstrong>Settings → Agents\u003C\u002Fstrong> on that connection to sign in again. Until then that member\nis passed over: new runs, retries and the blocked run itself go to another member of the role whose sign-in works.\nIf nobody's works, the run waits on \u003Cstrong>Blockers\u003C\u002Fstrong> and goes ahead as soon as one of them signs in again, and not\nbefore: there is no timer. A timeout, a rate limit or a server error while renewing a sign-in is not this: the run\ntries again on the usual backoff.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Blockers\u003C\u002Fstrong> lists a connection with a limit used up under \u003Cstrong>No LLM capacity\u003C\u002Fstrong>, once per connection: which limit,\nwhen it resets (in UTC), and \u003Cstrong>Raise limit\u003C\u002Fstrong> or \u003Cstrong>Add credits\u003C\u002Fstrong>. A key counts as used up only on what the provider\nreports as left on it, the same figure \u003Cstrong>Agents\u003C\u002Fstrong> shows, never on its all-time spend against a monthly limit.\u003C\u002Fp>\n\u003Cp>The OS also checks every ten minutes that the model each tier runs still exists on your connection. A tier whose own\nmodel is gone runs the \u003Ca href=\"#model-connections\">next best one\u003C\u002Fa> instead, so a model is listed on \u003Cstrong>Blockers\u003C\u002Fstrong>\nunder \u003Cstrong>Models unavailable\u003C\u002Fstrong> only when the provider says it is gone and its model list could not be read to choose\nanother. See \u003Ca href=\"\u002Fdocs\u002Fwork\u002Fruns#blocked-runs\">Runs\u003C\u002Fa>.\u003C\u002Fp>\n\u003Ch2 id=\"plans\">Plans\u003C\u002Fh2>\n\u003Cp>Your plan sets how long run history is kept (see \u003Ca href=\"\u002Fdocs\u002Fcontrol\u002Fisolation\">Isolation and retention\u003C\u002Fa>), and what\nyour enterprise includes: members, runs, runner minutes, concurrent runs, schedules and storage. See\n\u003Ca href=\"\u002Fpricing\">pricing\u003C\u002Fa>.\u003C\u002Fp>\n\u003Ch2 id=\"plan-limits\">Plan limits\u003C\u002Fh2>\n\u003Cp>The amounts your plan includes are held to. They sit beside your own \u003Ca href=\"#caps\">caps\u003C\u002Fa>, not instead of them: the\n\u003Cstrong>Spend\u003C\u002Fstrong> switch turns your caps off, never the plan's limits.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Where a plan says stop, work past it does not start.\u003C\u002Fstrong> Free stops at its portal runs, its workspace runner minutes\nand its storage. A run that would start past a stop is paused, not failed:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>It waits on \u003Cstrong>Blockers\u003C\u002Fstrong> under \u003Cstrong>Paused: plan limit\u003C\u002Fstrong>, naming the limit it met: for example, all 100 of Free's\nportal runs for the month used, which reads \u003Ccode>plan_limit: portal_run 100 of 100 on Free\u003C\u002Fcode>. \u003Cstrong>Needs you\u003C\u002Fstrong> shows the\nlimit with \u003Cstrong>Upgrade\u003C\u002Fstrong>, and the run's member says so in chat.\u003C\u002Fli>\n\u003Cli>It keeps its task and its member, like a run paused for spend.\u003C\u002Fli>\n\u003Cli>A run already under way when the limit is reached is not stopped. It finishes.\u003C\u002Fli>\n\u003Cli>A schedule that fires past a stop starts nothing, and leaves a skipped run that names the limit.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Paused work resumes by itself, within a minute of the limit clearing: when a new month begins (on Free, the 1st of\nthe month, UTC), when storage is freed below what the plan includes, or when the enterprise moves to a plan that\nincludes more. Nobody has to retry it.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Paid plans run on past the included amounts\u003C\u002Fstrong> and bill the overage per billing cycle on the renewal invoice: extra\nmembers, runner minutes, portal runs and storage, each at the plan's own rate. A billing cycle is a month counted from\nthe day the subscription started. A monthly plan is charged for it at the end of each cycle; an annual plan is still\ncounted month by month, and charged once a year, on the yearly renewal invoice. See\n\u003Ca href=\"\u002Fdocs\u002Fcompany\u002Fbilling#overage\">Billing\u003C\u002Fa>.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Members, schedules and concurrent runs are capped by the plan:\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cdiv class=\"prose__table\">\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Limit\u003C\u002Fth>\n\u003Cth>What happens at it\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Members\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>On Free, a member past the 2 included is refused. A paid plan adds them and bills the extra members.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Schedules\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>Enabling a schedule past the plan's number (3 on Free, 15 on Starter) is refused. Company and above have no limit.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Concurrent runs\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>The company's \u003Cstrong>Concurrent runs\u003C\u002Fstrong> can be set no higher than the plan's figure, plus any extra concurrent runs bought on \u003Cstrong>Billing\u003C\u002Fstrong>. A run past it waits its turn: it is not paused.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\n\u003C\u002Ftable>\n\u003C\u002Fdiv>\n\u003Cp>A refusal names the plan and says to upgrade on \u003Cstrong>Billing\u003C\u002Fstrong>.\u003C\u002Fp>\n\u003Cp>The OS's own tasks do not count toward portal runs: onboarding, self-improvement and the organisation changes take\nnone of the plan's portal runs, are never billed as extra ones, and are not held when the plan's portal runs are used\nup. One that works in a \u003Ca href=\"\u002Fdocs\u002Ftools\u002Fworkspaces\">workspace\u003C\u002Fa> still uses runner minutes. Their cadence takes none of the\nplan's schedules either.\u003C\u002Fp>\n\u003Ch2 id=\"over-the-api\">Over the API\u003C\u002Fh2>\n\u003Cdiv class=\"prose__table\">\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Call\u003C\u002Fth>\n\u003Cth>Scope\u003C\u002Fth>\n\u003Cth>What it does\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\n\u003Ctr>\n\u003Ctd>\u003Ccode>GET \u002Fv1\u002Fspend\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>spend:read\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>Today's spend and runs, the tightest caps, and every cap.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode>PATCH \u002Fv1\u002Fspend\u002Fcaps\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>spend:write\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>Add or replace the cap on one layer. Raising a day cap queues paused runs again.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode>PATCH \u002Fv1\u002Fspend\u002Fenabled\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>spend:write\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>Turn spend limits on or off: \u003Ccode>{ &quot;enabled&quot;: false }\u003C\u002Fcode>.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode>GET \u002Fv1\u002Fllm\u002Fmodels\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>llm:read\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>The models your connections offer, for a task's \u003Cstrong>Model\u003C\u002Fstrong>.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode>GET \u002Fv1\u002Fllm\u002Fconnections\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>\u003Ccode>llm:read\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>Your model connections, without their keys.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\n\u003C\u002Ftable>\n\u003C\u002Fdiv>\n",[12,16,19,22,26,29,32,36,39,42,45,48,51,54,57,60],{"id":13,"text":14,"level":15,"planned":8},"your-keys-your-bill","Your keys, your bill",2,{"id":17,"text":18,"level":15,"planned":8},"what-counts-as-spend","What counts as spend",{"id":20,"text":21,"level":15,"planned":8},"caps","Caps",{"id":23,"text":24,"level":25,"planned":8},"when-a-cap-is-reached","When a cap is reached",3,{"id":27,"text":28,"level":25,"planned":8},"per-run-budgets","Per-run budgets",{"id":30,"text":31,"level":25,"planned":8},"turning-spend-limits-off","Turning spend limits off",{"id":33,"text":34,"level":15,"planned":35},"caps-still-to-come","Caps still to come",true,{"id":37,"text":38,"level":15,"planned":8},"model-tiers","Model tiers",{"id":40,"text":41,"level":25,"planned":8},"which-model-a-run-uses","Which model a run uses",{"id":43,"text":44,"level":25,"planned":8},"which-model-a-chat-uses","Which model a chat uses",{"id":46,"text":47,"level":15,"planned":8},"model-connections","Model connections",{"id":49,"text":50,"level":25,"planned":8},"when-a-connection-does-not-offer-the-model","When a connection does not offer the model",{"id":52,"text":53,"level":25,"planned":8},"when-a-connection-cannot-be-used","When a connection cannot be used",{"id":55,"text":56,"level":15,"planned":8},"plans","Plans",{"id":58,"text":59,"level":15,"planned":8},"plan-limits","Plan limits",{"id":61,"text":62,"level":15,"planned":8},"over-the-api","Over the API",1794928142861]