目前暂不开放使用！
玄黄提供 API 中转服务，支持访问 GPT、Claude 及多种开源大语言模型。
支付方式
计费类型
https://apis.soys.site/1
https://apis.soys.site完成站点认证，解锁以下权益：
排名基于社区提交的测试数据与定期健康探测，仅供参考，非官方数据。
目前暂不开放使用！
按 16 个模型行对比最新 audit、最新测速、吞吐、延迟与按 token 计费价格。
| 模型 | 输入 ($/M) | 输出 ($/M) | 检测 | 速度 | 延迟 |
|---|---|---|---|---|---|
gemma-4-31b-多模态模型 | - | - | 727284100 | 1197.5 t/s | 0.23 s |
gpt-image-2cherry stuido生图 | $0.300/request | - | — | — | — |
gpt-image-2kelivo生图 | $0.300/request | - | — | — | — |
grok-4.20-agentgrok | $0.010/request | - | — | — | — |
grok-4.3-fastgrok | $1.25/M | $2.50/M | — | — | — |
| - | - | — | 47.5 t/s | 3.21 s | |
翻译/标题/OCR模型 | - | - | — | 260.7 t/s | 0.91 s |
| - | - | — | 32.9 t/s | 1.10 s | |
grok-4.1-fast | - | - | — | 73.0 t/s | 4.71 s |
glm-4.7-特别版 | - | - | 668465100 | 428.8 t/s | 3.12 s |
gemini-3-flash-preview-search | - | - | — | 179.5 t/s | 6.64 s |
| - | - | — | 43.2 t/s | 1.65 s | |
| - | - | — | 156.0 t/s | 4.01 s | |
| - | - | — | 1796.3 t/s | 0.49 s | |
| - | - | — | 244.5 t/s | 1.16 s | |
| - | - | — | 687.0 t/s | 1.62 s |
当前显示 16 / 16 个模型行
| 时间 | 模型 | 速度 | 延迟 |
|---|---|---|---|
| Jul 4, 05:21 PM | gemma-4-31b-多模态模型 | 1197.53 tok/s | 0.23s |
| Mar 31, 06:19 PM | qwen-3-235b | 686.99 tok/s | 1.62s |
| Mar 31, 06:19 PM | gpt-oss-120b-medium | 244.47 tok/s | 1.16s |
| Mar 31, 06:16 PM | gpt-5.3-codex | 47.55 tok/s | 3.21s |
| Feb 16, 03:43 AM | 翻译/标题/OCR模型 | 260.68 tok/s | 0.91s |
| Feb 16, 03:41 AM | claude-opus-4-6-thinking | 43.24 tok/s | 1.65s |
| Feb 16, 03:39 AM | kimi-k2-instruct | 32.90 tok/s | 1.10s |
| Feb 16, 03:26 AM | gpt-oss-120b | 1796.31 tok/s | 0.49s |
| Feb 16, 03:22 AM | grok-4.1-fast | 72.98 tok/s | 4.71s |
| Feb 16, 03:17 AM | gemini-3-flash | 155.96 tok/s | 4.01s |
用 325 个 LMSpeed 信号，把 玄黄 的 6 个相近 API 替代服务商放在一起比较：共享模型覆盖、价格、实测速度、可用性和免费模型。
| 服务商 | 对比理由 | 模型数 | 免费项 | 均价 | 速度 | 30 天可用性 |
|---|---|---|---|---|---|---|
| 玄黄 apis-soys-site Xuanhuang provides an API relay for accessing AI models including GPT, Claude, and open-source LLMs. | 当前服务商基线 | 48 | 0 | $0.621/M | 429 tok/s | 99.4% |
| Dext API ai-dext-top Dext API is an OpenAI-compatible API relay offering access to 500+ LLM models at competitive prices, with multi-channel routing and broad model coverage. |
| 343 | 228 | $111.69/M | 19 tok/s | 99.7% |
vsllm-com VSLLM runs a New API-powered AI gateway on vsllm.com for aggregated model access through a single endpoint. |
| 120 | 8 | $172.09/M | 184 tok/s | 99.7% |
api-6345ywz-cn 6345ywz API is an OpenAI-compatible API relay providing access to multiple AI models with competitive pricing. |
| 105 | 0 | $0.335/M | 241 tok/s | 99.7% |
naapi-cc Na API (naapi.cc) is an OpenAI-compatible LLM API gateway with competitive pricing and stable access to 100+ models from OpenAI, Anthropic, Google, and more. |
| 78 | 0 | $0.682/M | 649 tok/s | 99.4% |
tianxu-api Tianxu API provides an AI model relay service with multiple access points and stable connectivity. |
| 347 | 1 | $164.08/M | 93 tok/s | 99.1% |
llm-whitedream-top RenRen API runs a New API-powered gateway on llm.whitedream.top for aggregated access to multiple AI models. |
| 269 | 0 | $16.05/M | 271 tok/s | 99.5% |
