Most of this page is information other API resellers leave out. We would rather you learn it here than discover it later — including the parts that are inconvenient for us.
There is no magic in the middle. We are a gateway in front of the model makers' own APIs.
Three consequences worth being explicit about:
If retention matters to you, read the upstream policy for the model you plan to use — Alibaba Cloud Model Studio (Bailian), Moonshot, DeepSeek or Zhipu — and decide based on that, not on our summary of it.
Every price we charge — normal input, output, and cached input — is the vendor's own rate with the same discount applied. That means the cache maths you already know from going direct carries over unchanged: caching becomes worthwhile at exactly the same point it would with the vendor.
| Model | Vendor in (miss / hit) | gloritoken in (miss / hit) | Discount |
|---|---|---|---|
| Qwen 3.8 Max | $2.00 / $0.17 | $1.04 / $0.13 | ~48% / ~24% |
| Kimi K3 | $3.00 / $0.30 | $1.73 / $0.17 | ~42% / ~42% |
| GLM 5.2 | $1.40 / (see vendor) | $0.74 / $0.18 | ~47% |
Per 1M tokens. Vendor list rates as published, August 2026; explicit-cache creation fees follow the same rule.
The headline: Qwen 3.8 Max at $1.04 / $3.11 against $2 / $6 on both the official API and OpenRouter — the same model, the same official source, 48% less. You can verify every number here with cn-llm-bench, an open-source tool we maintain that measures us against everyone else and prints the results exactly as measured.
No vendor writes this section. We think that is why nobody trusts vendors. Do not use gloritoken if any of these apply:
If one of these is you, use the vendor's own international endpoint. That is the right answer, and we would rather say so than take money we should not have.
Still unsure whether this is a fit? Email hello@gloritoken.ai and describe your use case. If the answer is "you should go direct", that is what you will get.