Need to keep working right now? Codex credits ¥180 per 500 (1–20 packs, on your own Plus / Pro account, usually 5–30 minutes) · Pro 100 · break-even table
Update 2026-09-30: OpenAI reopened the $200 tier on 30 September as Pro 200 (new subscriptions get a lower allowance than before the pause), launched Pro 500 with Ultrafast and renamed the $100 tier Pro 100. Its pricing docs now say the Pro plans have no five-hour limit in Codex, only weekly limits. The upgrade advice and tables below reflect this; credits still extend usage on your current plan and are not an upgrade. Choosing a Pro tier: Pro 100 vs 200 vs 500.

Step one: which limit did you hit?

On Plus, Codex allowances run on two clocks: usage per 5-hour window, and a weekly limit that may apply on top. The Pro plans currently have no five-hour limit in Codex, so for them the weekly limit is the one that runs out. Local messages and cloud tasks share the same allowance, and turns that generate images use it 3–5× faster. The usage dashboard at chatgpt.com/codex/settings/usage, or /status in Codex CLI, shows which limit ran out, what's left and when it resets.

Two official details worth knowing: the active turn is allowed to finish (subject to fair use), so you don't lose a half-done change; and OpenAI Support doesn't reset limits — the options in the limit notice (add credits, use a reset you have, upgrade, wait) are all the options there are.

OpenAI's five ways forward

OptionCostBack to workBest forWatch out
Wait for the reset0At the reset time on the dashboardNo deadlineA weekly limit can mean days
Switch to a smaller model0ImmediatelyRoutine edits, testsOn Plus, GPT-6 Luna is about 350–3,000 messages per 5 hours vs 15–150 on GPT-6 Sol
Buy credits¥180 ≈ 26.83 USDT per 500Once delivered (usually 5–30 min)Occasional overruns; Plus users who want more GPT-6 AstraValid 12 months; not for Free accounts
UpgradePro 100 ¥850 ≈ 126.68 USDT / 30 daysOnce activated (usually 5–30 min)Cut off 3+ times a weekA new Pro 200 gets less allowance than before the pause; Ultrafast is Pro 500 only
Use an API keyAPI list pricesAs soon as the key is setCI, scripts, automationSeparate API account and payment method; no cloud features

A decision table by how often you hit the wall

How oftenWhat to doProduct
1–3 times a month, around deadlinesKeep 1–2 packs of credits as a buffer; what you don't use stays valid for 12 monthsCodex credits
Once or twice a weekSwitch models first (Sol / Luna for routine work, Astra for hard problems); add credits if that's not enoughCodex credits
3+ times a week, every dayUpgrade to Pro 100: more allowance than Plus (OpenAI said about 5× in mid-September), no five-hour limit in Codex (weekly limits apply), GPT-6 Pro in chatPro 100 on your account / ready-made
Still hitting weekly limits on Pro 100Pro 100 + credits; Pro 200 (new subscriptions reported at about 10× Plus); or Pro 500 (reported about 25×, plus Ultrafast)Pro 200 on your account / Pro 500
CI and batch scriptsAn API key, billed per use, outside your subscriptionWe don't sell API access

Break-even: Plus + credits, or Pro 100?

At today's prices (refreshed hourly from the pricing page), here is how many packs a month it takes before upgrading is cheaper:

Packs per monthPlus + creditsvs Pro 100 (¥850)
0 (0 credits)¥185 ≈ 27.58 USDT¥665 cheaper
1 (500 credits)¥365 ≈ 54.40 USDT¥485 cheaper
2 (1,000 credits)¥545 ≈ 81.23 USDT¥305 cheaper
3 (1,500 credits)¥725 ≈ 108.05 USDT¥125 cheaper
4 (2,000 credits)¥905 ≈ 134.88 USDT¥55 more
5 (2,500 credits)¥1,085 ≈ 161.70 USDT¥235 more
6 (3,000 credits)¥1,265 ≈ 188.53 USDT¥415 more

So: if you'd buy 4 or more packs every month, upgrade to Pro 100; at 3 or fewer, Plus + credits is cheaper. The same arithmetic one tier up: Pro 100 plus 3 packs a month costs more than Pro 200 on your own account (¥1,300) — keeping in mind that a new Pro 200 subscription gets a lower allowance than before the pause.

Two things the table leaves out, pulling in opposite directions:

  • Credits don't reset monthly. Unused plan allowance doesn't roll over; credits stay valid for 12 months. If your usage is spiky, buying credits as needed usually costs less than the table suggests.
  • Pro 100 isn't only more Codex. It has no five-hour limit in Codex (Plus gets about 5–45 GPT-6 Astra messages per 5 hours), and you can pick GPT-6 Pro in chat. If you use chat a lot, the upgrade is worth more than the table shows.

Why credits suit Plus users in particular

OpenAI's Codex pricing page says it directly: Plus includes limited GPT-6 Astra usage, and credits buy more. For many Plus users, day-to-day work fits in GPT-6 Sol (about 15–150 messages per 5 hours) or Luna; the bottleneck is the handful of hard problems that need Astra.

By OpenAI's rates a medium-sized task costs about 51 credits on GPT-6 Astra and about 10 on GPT-6 Sol. One 500-credit pack covers about 9 medium Astra tasks — for a few hard problems a week, far cheaper than paying ¥665 more every month for Pro 100. Every model and task size is in How far do 500 Codex credits go?

Switching models: the cheapest fix

It costs nothing and matters more than people expect. OpenAI's credit rates per 1M tokens, with relative cost for the same medium task (GPT-6 Sol = 1) and OpenAI's Plus estimate per 5 hours:

ModelInput / cached / outputRelative costPlus per 5 hours
GPT-6 Astra250 / 25 / 1,2505×5–45
GPT-6 Sol50 / 5 / 2501×15–150
GPT-6 Luna2.5 / 0.25 / 12.50.05×350–3,000
GPT-5.6 Sol100 / 10 / 5002×10–100
GPT-5.6 Terra50 / 5 / 3001.1×25–200
GPT-5.6 Luna5 / 0.5 / 300.11×250–2,000

GPT-5.5 was retired from Codex on 2026-10-14; if you relied on it, GPT-6 Sol is the natural replacement. Fast mode on the GPT-6 models bills purchased credits at 2× the Standard rate (2.5× against the plan's included allowance) — use it only when you're waiting on the result.

Is the API-key route worth it?

OpenAI lets anyone keep running local tasks with an API key at API prices. Per token it isn't cheaper: at the launch price of $40 per 1,000 credits, GPT-6 Astra in Codex costs 250 / 1,250 credits per 1M input / output tokens — exactly the API's $10 / $50 — and the API also bills cache writes ($12.50 per 1M for Astra), which Codex credit billing doesn't. API-key mode also loses Codex's cloud features (GitHub code review, Slack) and needs a separate OpenAI API account and payment method.

Verdict: API keys for CI and batch jobs; credits for a developer who just hit the wall. OpenAI is explicit that credits are not API credits — the two balances don't mix.

What to order on ClaudeMax

Everything is paid in USDT on TRC20 or BEP20. Top-ups usually land in 5–30 minutes; ready-made accounts are hand-delivered after you verify a delivery email (the order page shows 30 minutes to about 4 hours). Contact the bottom-right live chat promptly if anything goes wrong.