APINebula Review: Pricing & Comparison
Hong Kong-operated relay at ¥1 = $1 of credit that publishes a discount ratio per model, with a public 15%-max partner programme
Last verified: 2026-10-15 · Visit official site →
A Hong Kong relay that publishes a discount ratio per model
APINebula (apinebula.com) does not hide that it is a relay, while emphasising unified access to Anthropic and OpenAI models. Its pricing page is the most transparent of the vendors added in this round: every model carries an input price, an output price, a cache price and the resulting discount ratio, and the discount formula is documented on the fee page.
The billing rules themselves are simple:
- ¥1 of balance = $1 of model credit, ¥20 minimum, balance never expires, Alipay supported
- Overseas models carry an underlying ~1.43-discount FX subsidy (computed at 1 ÷ 7.0)
- Domestic models bill at official RMB list price (no FX subsidy)
- Combined discount = FX subsidy × group multiplier
Live examples (per 1M tokens):
- gpt-5.6-sol: ¥1.95 in / ¥11.70 out (0.58x)
- claude-opus-4-8: ¥2.50 in / ¥12.50 out (0.75x)
- claude-sonnet-4-6: ¥1.50 in / ¥7.50 out (0.75x)
- gpt-image-2: ¥0.05 in (1.49x)The console also exposes a public cost estimator (their own example: 100K input + 300K output = ¥6.24 per call), so you can price a workload before committing.
On promotions, it runs a partner programme with published numbers: up to 15% commission, split into a developer-ecosystem track (cash sponsorship or API credit) and a creator/KOL track, with 3-5 working-day approval. Relays rarely state a rebate percentage on the page, though the detailed settlement terms are not published.
Three things to keep in mind
First, the operating entity is a Hong Kong company, HONG KONG GALAXY VIDEO CO., LIMITED — the industry word in the name is video recording, which does not match an AI gateway. Hong Kong entities distributing API access are common enough, but this one is hard to verify independently.
Second, there is no public FAQ (visiting /zh/faq redirects to the login page) and the full model list is behind login, so the homepage counter claiming 56 actual models cannot be checked from outside.
Third, the latency claims contradict each other: the homepage says “99.9% availability SLA, latency under 200ms”, while the enterprise page says “under 50ms average gateway latency”. Both are vendor self-claims and they disagree, which makes them read more like marketing copy than measurements.
The price table also lists model IDs such as gpt-6-astra and gemini-4-argon. As of our own September 2026 verification of Google’s Gemini line-up, no Gemini 4 entry exists in the official model catalogue; relay listings do not always map one-to-one to official model IDs, so verify these names before integrating.
Who it’s for
If your priorities are “prices I can actually calculate, a low entry barrier, and a balance that never expires”, APINebula’s transparent pricing and ¥20 minimum make it worth testing — and it suits developers willing to double-check model IDs. If instead you need a long-term supplier with a clear legal entity and verifiable SLA, this vendor cannot provide that yet: validate with a small top-up and keep a second provider as backup.
Information verified 2026-10-15. Prices, discount ratios and the commission rate reflect what apinebula.com and its fee page displayed on that date; relay pricing and promotions change frequently.
Related reviews
- CatRouter: a mainland-direct aggregator at ¥1 = $1 with published group multipliers, one key spanning OpenAI/Anthropic/Gemini native protocols
- CCTQ: runs a New API panel and publishes per-group measured availability (83.6% average) plus a no-questions refund policy
- PoloAPI: balances stability and multi-model coverage, focused on usage statistics and multi-project call auditing
- AiHubMix: developer-integration-friendly, permanently free test tier, Prompt Caching support, thorough documentation
Quick facts
| Pricing model | ¥1 of balance = $1 of model credit, ¥20 minimum top-up, balance never expires, Alipay supported; overseas models carry an underlying ~1.43-discount exchange-rate subsidy (computed at 1 ÷ 7.0), domestic models bill at official RMB list price, and the combined discount = FX subsidy × group multiplier; e.g. gpt-5.6-sol ¥1.95 in / ¥11.70 out (0.58x), claude-opus-4-8 ¥2.50 in / ¥12.50 out (0.75x) |
|---|---|
| Model coverage | The homepage counter states 56 actual models, and the pricing page lists each model with input/output/cache prices; OpenAI, Anthropic Claude, Google and image models are included, with Grok and GLM/Kimi CLI pages in the docs; the full list requires login |
| Latency / SLA | The homepage claims 99.9% SLA with latency under 200ms, while the enterprise page states under 50ms average gateway latency and 10M+ daily requests — two mutually inconsistent latency figures on the same site, both vendor self-claims with no third-party verification |
| Mainland direct connect | Claims stable direct connectivity and Alipay top-ups without an overseas card (it does not state mainland direct-connect verbatim; test on your own network) |
| Best for | Developers / Enterprise |
| Referral program | The partner programme publishes a maximum 15% commission split across a developer-ecosystem track (cash sponsorship or API credit) and a creator/KOL track, with 3-5 working-day approval; the detailed settlement terms are not published. |
Pros
- The best pricing transparency of the vendors added in this round: every model shows input, output and cache prices plus the resulting discount ratio, the fee page documents the discount formula (FX subsidy × group multiplier), and the console has a public cost estimator
- A partner programme with a published rate — up to 15% commission across two tracks — which is rare among relays, most of which never state a number
- Low commitment: ¥20 minimum top-up, balance never expires, Alipay and VAT invoices supported, and an enterprise path with corporate transfer, contracts and invoicing
Cons
- The operating entity is a Hong Kong company, HONG KONG GALAXY VIDEO CO., LIMITED — a name whose industry is video recording, not AI gateways, and which is hard to verify independently
- There is no public FAQ (/zh/faq bounces to the login page) and the full model list is behind login, so the homepage's 56-model counter cannot be checked from outside
- The latency claims contradict each other (under 200ms on the homepage, under 50ms on the enterprise page), and the price table lists model IDs such as gpt-6-astra and gemini-4-argon that do not appear in the official catalogues — verify before integrating
Compare more AI API relays
See the full comparison board — filter by price tier, model coverage, and mainland direct-connect status.
Back to the comparison board →