API Compass tracks and independently measures relay and aggregator services worldwide, publishing each one's uptime, response latency, model coverage and backend engine — with the service's own advertised rates and minimums listed separately. We do not sell, resell or proxy anything; we just lay the data out.
Work out what you actually need first, then choose — that beats starting from a leaderboard.
Services with strong scores and uptime, and reasonably transparent information. A listing is not an endorsement — start with a small top-up.
Per the service's own site: Fast AI inference billed per token. DeepSeek V3.2, Kimi K2.5, GLM-5.…
Ranked by overall score; ties broken by higher uptime.
| # | Service | Verification | Models supported | Uptime | Latency | Models | Score |
|---|---|---|---|---|---|---|---|
| 1 | Avian avian.io |
Verified | Claude · GPT · DeepSeek · Llama | 100% | 1438 ms | 19 | 9.9 |
| 2 | Featherless featherless.ai |
Verified | GPT · DeepSeek · Qwen · Llama · Mistral | 100% | 1274 ms | — | 9.8 |
| 3 | TiMi CC timicc.com |
Verified | GPT · Qwen | 100% | 1727 ms | — | 9.8 |
| 4 | Yuegle API api.yuegle.com |
Verified | Claude · GPT · Gemini · Grok · DeepSeek | 100% | 603 ms | — | 9.8 |
| 5 | api.gpt.ge api.gpt.ge |
Verified | Claude · GPT · Gemini · Grok · DeepSeek · Qwen | 100% | 450 ms | — | 9.8 |
| 6 | api.vveai.com api.vveai.com |
Verified | Claude · GPT · Gemini · Grok · DeepSeek · Qwen | 100% | 412 ms | — | 9.8 |
| 7 | ePhone AI api.ephone.ai |
Verified | Claude · GPT · Gemini · DeepSeek | 100% | 1428 ms | — | 9.8 |
| 8 | jeniya.cn jeniya.cn |
Verified | Claude · GPT · DeepSeek | 100% | 2120 ms | — | 9.8 |
| 9 | apipro.maynor1024.live apipro.maynor1024.live |
Verified | Claude · GPT · Gemini | 98.36% | 1561 ms | — | 9.8 |
| 10 | 便携AI聚合API api.bianxieai.com |
Verified | Claude · GPT · Gemini · Grok · DeepSeek | 98.36% | 786 ms | — | 9.8 |
Figures below come from information the services published themselves (source noted on each page). A rate of 1.0 means official list price; below 1.0 is below list price.
| Service | Minimum top-up | Billing | GPT | Claude | Gemini | Notes |
|---|---|---|---|---|---|---|
| 犀速AI xisu.ai |
¥50 | Pay-as-you-go | ×0.1 | ×0.3 | ×0.1 | Site states: Claude at 0.3x, GPT / Gemini / Grok at 0.1x… |
| Lever AI leverai.net |
Not published | Pay-as-you-go / subscription | ×0.35 | — | — | The service advertises a 0.35 rate on its standard GPT c… |
| AI302 Api (api.302ai.com) api.302ai.com |
$1 | Pay-as-you-go | — | — | — | |
| AI302 Api (api.302ai.cn) api.302ai.cn |
$1 | Pay-as-you-go | — | — | — | |
| OpenRouter openrouter.ai |
$5 | Pay-as-you-go | — | — | — | |
| 稳明光语纪 wenming7.cn |
¥1 | Pay-as-you-go / subscription | — | — | — | The service advertises pay-as-you-go from ¥1 and subscri… |
| 百川汇智API rivernexus.vip |
¥1 | — | — | — | — | |
| 维塔斯中转站 www.aiwts.com |
10 | — | — | — | — | |
| tbtk中转站 tbtk.asia |
5 | — | — | — | — | |
| nimblecut(剪灵) nimblecut.com |
9.9 | — | — | — | — |
Average time to reach the service's home page — a useful first indicator of network quality.
| # | Service | Response latency | Stability | Uptime | Score |
|---|---|---|---|---|---|
| 1 | 灵绘 AI | 22 ms | Very stable | 100% | — |
| 2 | 稳明光语纪 | 49 ms | Very stable | 100% | 5.5 |
| 3 | chatfire.cn | 63 ms | Very stable | 100% | 9.3 |
| 4 | 七牛云开发者平台 | 64 ms | Very stable | 100% | 6.3 |
| 5 | 智谱丨BigModel 平台 | 66 ms | Very stable | 100% | 6.3 |
| 6 | foxaigc.com | 103 ms | Stable | 98.36% | 6.3 |
Decide which model you need to call first, then see who covers it.
Ten minutes well spent before you top up.
The relay industry's problem is not a shortage of directories. It is that nearly every directory copies its da…
This wave of model names mixes generations, product codenames, tier suffixes and task markers. Read them as ve…
You open your laptop, get 404 model_not_found, and production is down. It is almost always an upstream retirem…
Rate limiting does not always return 429. Silent queueing, silent downgrades and mid-stream cutoffs report no…
The biggest problem in this market is information asymmetry: every landing page claims to be stable and cheap, but real uptime, actual response speed, whether a given model is genuinely available, and how a quoted rate is calculated are all impossible to check before paying. And mistakes are expensive — prepay for credit, the service disappears or quietly downgrades its models, and the money is gone.
So we do three things:
We do not proxy, resell or top up APIs, take no part in any transaction, and do not vouch for any service's finances. See About for how listing and ranking work.