API2D Review: Pricing & Comparison
A veteran paste-your-key-and-go relay with broad client compatibility, priced at roughly 1.5x official rates
Last verified: 2026-10-15 · Visit official site →
A relay that is still running but priced in the past
API2D (api2d.com) is one of the earliest Chinese relays built around “register, top up, paste your key”, positioning itself as “using an API like using electricity”, and it states plainly that it is not a pure relay: “API2D is not simply an API relay — we also have a large amount of server-side enhancement (knowledge bases, search, business plugin APIs) and an agent/reseller feature set.”
Its API is OpenAI-compatible (base_url https://oa.api2d.net/v1/chat/), the site and FAQ load normally, and the signup/top-up flow works — so to be clear up front: this is a service that is genuinely still operating.
The problem is equally clear: its pricing is stuck at a market position that no longer exists.
Point (P) billing, per the site:
- “Overall pricing is held at 1.5x the upstream provider's official price (except the ¥3.5 tier)”
- 1P = 133.3 GPT-3.5 (non-16k, non-1106) input tokens; minimum 1P per call
- 24-hour response-cache hits are not billed extra (bypass with the x-api2d-no-cache: 1 header)
- Points expire 180 days after your last top-upWhat does 1.5x official mean in practice? The other Chinese relays added in this round advertise 0.15x–0.75x of official pricing, which puts API2D’s effective unit price at roughly 2–10x that of its peers. This is not to say the service is absurdly expensive — only that its pricing logic still reflects the era when official account quota was scarce and reselling at a markup was the norm.
Coverage: no 2026 models in sight
The model families listed are most OpenAI endpoints, Claude, Stable Diffusion, Azure TTS, vector databases and Bing Tools. Gemini, DeepSeek, Qwen, Kimi, GLM and Grok appear nowhere on the site; the API documentation still uses gpt-3.5-turbo in its examples and was last updated on 2025-11-17.
For long-time users that combination still works. For new users, missing Gemini and the entire domestic model camp is a real coverage gap.
Who API2D still suits
Honestly, in 2026 it suits two groups: existing account holders with small workloads who cannot be bothered to migrate (mind the 180-day expiry — do not leave a large balance), and anyone who needs one key to get an older OpenAI-only tool working quickly. For value, current models or large-scale production use, the alternatives linked below are clearly better fits.
It also has no ordinary referral rebate, but it does have a reseller tier: once your balance exceeds 200,000P (about ¥400) you can open a sub-selling page and set your own prices — that is designed for people reselling credit, not for ordinary users.
Information verified 2026-10-15. Billing rules, point expiry and the model list reflect what api2d.com and its FAQ displayed on that date; its API documentation was last updated 2025-11-17.
Related reviews
- OpenRouter: a cross-vendor multi-model aggregation platform, 200-300+ models, one API key to call OpenAI/Anthropic/Google/Meta/Mistral and more
- YunWu API: 500+ model aggregation, ¥0.5/USD exchange rate, free daily GPT-4o calls via GitHub login
- AiHubMix: developer-integration-friendly, permanently free test tier, Prompt Caching support, thorough documentation
- TokenRiver: enterprise-grade AI gateway, 650+ models, ¥1=$1 exchange-rate advantage, multi-channel automatic failover
Quick facts
| Pricing model | Point-based (P) billing; the site states that overall pricing is held at 1.5x the upstream provider's official price (except the ¥3.5 top-up tier); 1P equals 133.3 GPT-3.5 input tokens and the minimum per call is 1P; 24-hour response-cache hits are not billed extra; points expire 180 days after your last top-up |
|---|---|
| Model coverage | Most OpenAI endpoints, Claude, Stable Diffusion, Azure TTS, vector databases and Bing Tools; there is no Gemini, DeepSeek, Qwen/Kimi/GLM or Grok entry anywhere on the site, and the API documentation still uses gpt-3.5-turbo as its example — the model line-up is visibly dated |
| Latency / SLA | Neither the site nor its FAQ publishes any latency or SLA figure; the API documentation was last updated on 2025-11-17 |
| Mainland direct connect | Mainland direct-connect unverified (no domestic-node claim on the site, and it sits behind Cloudflare) |
| Best for | Developers / Fallback / backup |
| Referral program | No ordinary referral rebate, but there is a reseller/distribution tier: once your balance exceeds 200,000P (about ¥400) you can open a sub-selling page and set your own prices. |
Pros
- One of the longest-running Chinese relays still standing: the site, signup and top-up flows all work, and the FAQ documents the billing rules fairly completely
- Strong client compatibility: the site lists dozens of supported apps (Chat酱, OpenCat, VSCode, Immersive Translate and more), which suits users who would rather not touch code
- Free 24-hour response caching and a 1P minimum per call keep light chat usage cheap, and the rules are stated in the FAQ rather than hidden
Cons
- At roughly 1.5x official pricing it is uncompetitive in a 2026 market where peers advertise 0.15x–0.75x of official rates
- Points expire 180 days after your last top-up, so an idle balance is simply lost
- Model coverage is clearly dated: no Gemini and no domestic models such as DeepSeek, with the API documentation last updated 2025-11-17
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 →