An OpenModels account carries two balances that behave nothing like each other. One refills on a clock and cannot be bought; the other is bought and never refills on its own. Almost every question about pricing here resolves once you know which wallet is paying, so this page is the long version: what fills each one, what each one is allowed to spend on, and the three places the simple rule has an exception.
Green credits (⚡) refill on a clock and pay for models whose work happens on hardware we own. Gold credits (✦) are bought with money and pay for anything in the catalogue, including the hosted models from Google, ByteDance, OpenAI and the rest. You cannot convert one into the other, and green is never sold.
The split is not a marketing device, it follows the bill. A generation on our own GPU costs electricity on a machine that is already paid for, so it can refill for nothing. A generation on a hosted API costs real dollars the moment it is called, so it cannot.
Which plan: Every signed-in account has both wallets. What the plan decides is the size of the green cap, the size of the monthly gold grant, and which models you are allowed to select at all. Signing up grants zero gold — a brand new account has a green wallet and nothing else.
Green tops itself back up continuously toward your plan's daily cap, at one twenty-fourth of the cap an hour. On a free account that is ⚡20 a day, so roughly one credit every 72 minutes; on Creator it is ⚡200 a day, a little over eight an hour. There is no midnight reset to wait for — spend the lot at nine in the morning and the wallet is usable again within the hour.
Green pays for the models that run here: the own-GPU image lanes, the own-GPU video lanes, prompt enhance, and Describe on Aeon. Two exceptions are worth knowing before you plan a week's work around it.
What it costs: Own-GPU work starts at ⚡1 an image. Local prices are set at runtime and can carry a demand multiplier when the render queue is deep, so treat every own-GPU number on this site as a floor rather than a fixed price. The composer quotes the real figure on the Generate button before you press it.
| Plan | Price a month (annual) | ⚡ green a day | ✦ gold a month | Parallel jobs | Storage |
|---|---|---|---|---|---|
| Free | $0 | 20 | none | 2 | 50 MB |
| Starter | $7.50 ($5.00) | 100 | 50 | 3 | 250 MB |
| Creator | $49.99 ($34.99) | 200 | 380 | 5 | 2 GB |
| Pro | $100 ($75) | 400 | 900 | 10 | 10 GB |
| Ultra | $1,000 ($750) | 2,000 | 10,000 | 100 | 100 GB |
Storage is the quiet limit on that table. A single upload may use at most 90% of the plan's whole allowance, which is what really caps the file size you can bring in: 45 MB on Free, 225 MB on Starter, about 1.8 GB on Creator.
Gold arrives two ways: as the monthly grant that comes with a plan, or as a one-off bundle bought on the pricing page. Unlike green it accumulates rather than filling toward a ceiling, and nothing expires it — an unspent bundle is still sitting there next year.
There are five bundles: 100, 500, 1,000, 2,500 and 5,000 credits. The price per credit improves as the bundle gets bigger, and it improves again with the plan you are already on, so the same 1,000 credits cost a Creator less than they cost a free account. Both are on the same page as the plans, under #credit-packs.
Watch out: Buying credits does not change your plan. Model access is gated on the tier, not on the balance, so a free account holding 5,000 gold still cannot select a Starter-and-up model. If a specific model is what you want, the plan is the thing to buy; credits are for running more of what you can already reach.
Every composer quotes a price on the Generate button before you press it, and that quote is what you are charged. The shape of it: own-GPU images start at ⚡1; own-GPU video is billed by block of length, or per second of source on the clean-up lanes; hosted models are billed per call in gold. One hosted model is the exception — GPT Image 2 costs 0 credits, subject to a daily per-account limit and to the shared subscription's own availability.
Priority is a purchase, not a plan perk. Any account can buy a faster lane on a job-by-job basis, paid from either wallet — green for own-GPU work, gold for the rest.
| Lane | Multiplier | What it buys |
|---|---|---|
| Normal | ×1 | The default. First in, first out within the lane. |
| Fast | ×2 | Jumps ahead of everything queued in Normal. |
| Express | ×4 | Jumps ahead of everything below it. |
Those multipliers are the static floor. Each is surged by how deep the priced queue already is, with no ceiling, which is the honest way to price one shared GPU. Whatever the number is at submit is snapshotted onto the job, so the charge equals the quote you accepted, and a job already sitting in the queue can be boosted after the fact from its own card.
The invite panel on your profile hands you a personal link. It pays in green, on two rungs, and never in gold.
Chat is not priced in credits. Each family carries a daily allowance instead: the OpenAI models share a dollar-a-day budget that rises with the plan, the Claude models share a pool of replies. When an allowance runs out the app buys more out of your wallet, green first, rather than refusing the message — so chat reaches the wallet indirectly, without any reply having a credit price.
Video Analyze is billed entirely in gold. Green cannot pay for any part of it, frame analysis or moment search. It is worth planning for, because a long source at default sampling can be a meaningful share of a Starter month's grant in one run.
No. Green is never for sale, at any price. The only two things that raise it are the plan, which sets the daily cap, and invites, which pay green directly.
Green fills toward its daily cap and stops, so there is nothing to expire — but nothing carries over either. Gold accumulates and has no expiry; an unspent bundle stays on the account. Neither wallet is clawed back if you downgrade.
No. Credits are a wallet, not a plan. Access to a model is decided by your tier, so a free account with a large gold balance still cannot select a Starter-and-up model. Buying credits deliberately does not change your tier.
Roughly twenty on the own-GPU lane, where images start at ⚡1. Local prices are set at runtime and can carry a demand multiplier when the queue is deep, so treat it as about twenty rather than exactly twenty.
Green, while the model is running on our own GPUs. If the local lane is unavailable and the same model is served over a paid API instead, that run is billed in gold — the free wallet is not allowed to pay a bill with real dollars behind it.