Pro 20x has a weekly credit number you can watch. Claude does not.

No. Codex switched to a token-based credit balance on April 2, 2026, so you see a number, you watch it tick down, and you can buy more without changing plans. Claude Pro and Max 20x have no credit balance at all. Anthropic meters you with utilization fractions on a rolling 5-hour window plus several 7-day (weekly) windows, and bills overflow as pay-as-you-go 'extra usage'. There is no 'credits remaining' figure anywhere in the Claude apps. The closest analog is the percentage bar at claude.ai/settings/usage, and even that does not show every weekly bucket the server enforces.

Both ecosystems shipped a 20x tier, which is exactly why this comparison is confusing. ChatGPT Pro 20x is the $200/month plan, 20 times the Plus limits, roughly 200 to 1,200 Codex cloud tasks per 5-hour window. Claude Max 20x is also $200/month and is sized at about 20 times Claude Pro. The shared '20x' label hides a structural difference: Codex 20x meters you in credits you can see and buy, Claude Max 20x meters you in opaque utilization windows you cannot.

At least six. The /api/oauth/usage payload that claude-meter reads deserializes into a UsageResponse struct (src/models.rs) with seven_day, seven_day_sonnet, seven_day_opus, seven_day_oauth_apps, seven_day_omelette, and seven_day_cowork, alongside the five_hour window and an extra_usage block. Each seven_day* field is a Window object with a utilization float and a resets_at timestamp. Claude's own Settings page renders only a subset of these, so you can be fine on the headline weekly bar while a model-specific bucket like seven_day_opus is the one about to 429 you.

In one undocumented endpoint: GET https://api.anthropic.com/api/oauth/usage, authorized with the Bearer token Claude Code already stores in your macOS Keychain under the service name 'Claude Code-credentials'. That single response carries the 5-hour window, all the seven_day* windows, and the extra_usage block. claude.ai/settings/usage calls the same internal data and renders a trimmed version. claude-meter reads the raw response and shows every window in the menu bar.

No, and this is the most common mix-up. ccusage and Claude-Code-Usage-Monitor read ~/.claude/projects/**/*.jsonl on your disk and sum tokens against a price card. That is an accurate local token count, but it is not the server's utilization. The weekly buckets fold in peak-hour weighting, claude.ai web traffic, other devices on your account, and OAuth-app calls, none of which appear in your local files. They answer 'what did Claude Code burn here', not 'how close is Anthropic to weekly-capping me'. Only the server number tells you that.

Sort of, but it is not a credit purchase. When you cross a window, Anthropic can route overflow into metered 'extra usage', a pay-as-you-go dollar balance, rather than hard-stopping you. claude-meter surfaces that as the extra_usage block in dollars. The difference from Codex is that you do not pre-buy a credit pack and watch it deplete; you accrue overage after the fact, which is exactly why seeing the weekly utilization climb before you hit it matters more on Claude than on Codex.

It is free, open-source (MIT), and macOS 12 or newer. The menu bar app reads the OAuth token that the Claude Code CLI stores in your Keychain, so you need Claude Code installed and logged in (run 'claude' once). There is also a browser extension path that forwards your claude.ai session instead. Either way there is no manual cookie paste and no separate login. Install: brew install --cask m13v/tap/claude-meter.