Plan capacity
Set and reset estimated weekly and monthly API-dollar capacity.
Plan capacity is at the bottom of Settings → Agents → Claude or Codex. Expand Plan capacity (advanced) to edit Weekly or Monthly dollars. Changes save on blur on this device. Reset removes your override and returns to a locally learned estimate, or the default estimate when there is not enough history.
These figures estimate the API list-price value of included usage. They are not subscription prices, credits, invoices, or guaranteed provider allowances. Changing them does not change provider limits or reset times. Monthly is weekly multiplied by 30.4 / 7, not a separate monthly quota.
Where the defaults come from
Settings fields use your saved override first, then a capacity learned from local usage and weekly-limit changes, then the estimates below. Unknown plans remain blank until learned or edited. Claude reports preserve the 5x/20x tier from credential metadata when available, as display metadata, without changing historical subscription labels or IDs. Older unqualified “Max” readings use the conservative $500 seed; that fallback does not identify the account as Max 5x. Model-specific learned limits remain separate from the all-model weekly estimate.
| Plan | Weekly API-dollar estimate | Basis |
|---|---|---|
| Claude Pro | $100 | Seed inferred from the $500 Max 5x seed, scaled by the $20/$100 subscription-price ratio. Not a measured Pro weekly allowance. |
| Claude Max (tier unavailable) | $500 | Conservative seed until a precise tier or learned value arrives. |
| Claude Max 5x | $500 | Rounded down from a published $523 measurement. |
| Claude Max 20x | $1,100 | Published measurement. |
| Codex Plus | $120 | Seed inferred from the $600 Prolite measurement divided by 5. |
| Codex Pro / Prolite / Pro $100 | $600 | Published user measurement. |
| Codex Pro $200 | $2,400 | Seed inferred from $600 × 20/5. |
Unqualified Codex “Pro” conservatively assumes the $100 tier, matching the existing usage-borrowing assumption. This does not establish the actual subscription tier. These defaults fill Settings fields only. They do not change historical graphs, measured ceilings, or lending between accounts. Saved overrides still affect usage estimates.
The Botfarm measurement used 80 tasks over February 27–March 1, 2026, primarily Opus 4.6. Its author calls the figures lower bounds because interactive usage was not fully counted. The Codex Prolite measurement used more than 20 usage-percent/API-cost pairs on July 23, 2026 and reported about $600, down from $675 ten days earlier. It is one user's measurement, not OpenAI's.
OpenAI's pricing documentation describes Pro usage tiers relative to Plus; applying those ratios to API dollars is an AgentGrid seed assumption, not a published weekly dollar entitlement. Anthropic's Max documentation describes session usage multipliers and separate weekly limits. AgentGrid does not apply the 20x session multiplier to Claude's weekly estimate.
Sources checked October 16, 2026. These seeds can become stale as models, pricing, caching, workloads and provider policies change. Default estimate, Learned estimate, and Your override identify which value you are seeing. Prefer your own learned value or an override when a seed does not fit your account.
Which accounts appear
Plan capacity and the Usage account count omit removed accounts and identities currently signed out. An exhausted account is still signed in and remains visible. Removing an account does not erase its saved usage, snapshots, or overrides; historical spend remains in the Usage totals.