Usage Rebate Rules: How BazaarLink Milestone Credits Work

Query date: 2026-09-15. The figures in this article come from the settings currently in effect on the platform; the eligible model list and rules are authoritative on the usage rebate explanation page.

API users always wonder about one thing: if usage is high, can the unit price be lower? The original providers usually answer "contact sales for enterprise plans." BazaarLink has turned this into a public rule: Usage Rebate. Each time paid spending on a given plan in a month crosses a milestone, a fixed amount is credited to your balance. No application, no negotiation, and no waiting for a sales reply.

This article covers five things: how it is calculated, why the actual rate is not fixed, which spending does not count, when you receive it, and the situations in which you receive nothing at all.

How it works: fixed milestone rebates, not a discount

First, the most common misunderstanding: this is not "everything at 10% off." It is a set of milestones; each one crossed triggers a fixed amount, and all reached milestones add up. It does not only give the highest tier. Spending beyond the last milestone ($500) earns a further rebate at a tail percentage.

There are currently three plans:

Example: if you spend $120 in a month on the OpenAI GPT plan, you cross the $20, $50, and $100 milestones, so the rebate = $2 + $3 + $5 = $10. If you spend $1,000, all five milestones are reached ($50), plus $500 × 10% for the portion above $500 ($50), for a rebate of $100.

Why the actual rate is not fixed

Because the milestones are fixed amounts, how much you spend determines whether you land exactly on a milestone or fall between two of them. Landing exactly on a milestone is the best deal; in between, the rate is diluted:

So the correct understanding is: stable at 10% above $500 (DeepSeek 20%); between 4% and 10% below $500; zero below $20. If your monthly bill is in single-digit dollars, this mechanism does not matter for you; compare unit prices directly.

What counts and what does not

Counts: paid usage billed through the platform and actually deducted from your balance, where the model is on the plan's eligible list.

Does not count:

• Free-plan usage (for example, free models such as auto:free)
• Bring-your-own-key (BYOK) usage: that money goes to your own upstream provider, and the platform records a cost of 0
• Usage offset by subscription credits
• Models not on the plan list. Matching uses exact model IDs, not prefix matching. For example, the OpenAI GPT plan currently excludes the -pro variants as well as embedding, image, and video models; the DeepSeek plan currently only includes the V4 Flash and V4 Pro series, and R1, V3.2, and V4.1 Flash are not included. Please use the list on the explanation page as authoritative; do not assume "it's all the same company."

In addition, each plan is calculated separately. If you spend $15 on GPT and $15 on Gemini, and neither reaches $20, the rebate is 0; they are not combined into $30. Concentrating traffic from one provider family makes it easier to hit a milestone than spreading it across several providers.

When you receive it, and what you receive

• Settled by UTC calendar month; on the 1st of each month, the previous month's rebate is credited
• Recalculated every month: spending that does not reach a milestone in the current month does not carry over to the next month
• Credited to the wallet that generated the usage: personal usage goes to the personal wallet, and organization usage goes to the organization wallet
• The rebate is a balance that can be spent directly; it is not refunded, not separately invoiced, and does not expire. Your original paid top-ups still receive the standard unified invoice, and the rebate is simply an additional layer
• All three plans currently have no monthly cap

After logging in, the Credits page shows this month's eligible usage and estimated rebate.

Three real-world scenarios, including one that earns nothing

Applying the scenarios from our pricing article, using the same set of assumptions:

Two of the DeepSeek scenarios earn nothing, and that needs to be stated clearly. The DeepSeek rebate rate is the highest of the three plans (20%), but its unit price is also the lowest. Precisely because it is cheap, usage at a typical scale rarely crosses $20. DeepSeek's 20% is meant for high-traffic teams.

What this means for finance and procurement

• Predictable: the milestones are public rules, so you can include them in budgeting directly without waiting for negotiation
• No process change: TWD pricing, unified invoices, key spending caps, and organization cost allocation all stay the same; the rebate happens at the billing layer, and API calls do not change at all
• Do not spend more to earn the rebate: its value is getting part of the spending you were already going to make back. Spending an extra $1 to get back $0.10 is always a loss

Related

Unit prices and TWD estimates by provider: GPT, Gemini, DeepSeek. Next step after your free credits run out: How to get free AI API credits.

TWD billing · Taiwan invoices · leading AI models · OpenAI-compatible API