Tokeness gives developers one entry to every major AI model. Use one OpenAI-compatible key for GPT, Claude, DeepSeek and more, with quota control, routing management, usage audit and privacy-preserving relay.
- 延迟
- 0.00 s
- 创建时间
- 5/22/2026
- 充值汇率
- ¥7.00 / 1 美元额度
功能
任务数据导出签到
登录方式
GitHubPasskey
付费信息
支付方式
微信支付支付宝
计费类型
按量计费
- 退款
- 支持
- 开票
- 支持
API 地址
- 接口 1历史推断 / 未验证
https://tokeness.io - 接口 2历史推断 / 未验证
https://n.tokeness.io - 接口 3历史推断 / 未验证
https://n-cf.tokeness.io - 接口 4历史推断 / 未验证
https://tokeness.cn - 接口 5历史推断 / 未验证
https://n.tokeness.cn - <tnt l="zh">主节点</tnt><tnt l="en">Primary</tnt>
https://n.tokeness.dev/v1<tnt l="zh">全球加速，长超时，腾讯云CDN，Claude 要删尾部/v1</tnt><tnt l="en">Global acceleration, extended timeout, Tencent Cloud CDN</tnt>
- <tnt l="zh">备用</tnt><tnt l="en">Backup</tnt>
https://n-cf.tokeness.dev/v1<tnt l="zh">CloudFlare 全球加速，Claude 要删尾部/v1</tnt><tnt l="en">CloudFlare global acceleration</tnt>
- 接口 8历史推断 / 未验证
https://n.tokeness.dev - 接口 9历史推断 / 未验证
https://n-cf.tokeness.dev
认领该站点
I由 Im358 Buard 认证
排行榜排名
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。
关于 Tokeness.io
健康检查
99%近期可用性
历史记录 (72 条)
过去现在
API 基准测试与价格
按 74 个模型行对比最新 audit、最新测速、吞吐、延迟与按 token 计费价格。
gemini-3.7-flash | - | - | ||||||||||
glm-5.3 | - | - | ||||||||||
grok-4.6 | - | - | ||||||||||
qwen3.8-max-free | - | - | ||||||||||
qwen3.8-max | - | - | ||||||||||
kimi-k3-free | - | - | ||||||||||
claude-opus-5Claude-Entry | $0.300/M | $1.50/M | ||||||||||
claude-opus-5Claude | $1.25/M | $6.25/M | ||||||||||
claude-opus-5Claude-Official | $5.00/M | $25.00/M | ||||||||||
gemini-3.6-flashGemini | $0.450/M | $2.25/M | ||||||||||
gemini-3.5-flash-liteGemini | $0.090/M | $0.750/M | ||||||||||
gpt-5.5-pro | - | - | ||||||||||
kimi-k3default | $1.80/M | $9.00/M | ||||||||||
gpt-5.6-lunaGPT-Entry | $0.030/M | $0.180/M | ||||||||||
gpt-5.6-lunaGPT | $0.075/M | $0.450/M | ||||||||||
gpt-5.6-lunaGPT-Official | $1.00/M | $6.00/M | ||||||||||
gpt-5.6-terraGPT-Entry | $0.075/M | $0.450/M | ||||||||||
| $0.188/M | $1.13/M | |||||||||||
gpt-5.6-terraGPT-Official | $2.50/M | $15.00/M | ||||||||||
gpt-5.6-solGPT-Entry | $0.150/M | $0.900/M | ||||||||||
gpt-5.6-solGPT | $0.375/M | $2.25/M | ||||||||||
gpt-5.6-solGPT-Official | $5.00/M | $30.00/M | ||||||||||
grok-4.5Grok | $0.150/M | $0.450/M | ||||||||||
claude-sonnet-5Claude-Entry | $0.120/M | $0.600/M | ||||||||||
claude-sonnet-5Claude | $0.500/M | $2.50/M |
当前显示 25 / 74 个模型行
最近测试记录
| 时间 | 模型 | 速度 | 延迟 |
|---|---|---|---|
| Jun 29, 07:28 AM | tokeness/free | 79.17 tok/s | 1.52s |
| Jun 29, 04:31 AM | mimo-v2.5-pro | 50.64 tok/s | 1.75s |
| Jun 27, 01:54 PM | gpt-5.5 | 67.28 tok/s | 1.83s |
| Jun 27, 11:03 AM | gpt-5.5 | 33.02 tok/s | 6.54s |
| Jun 27, 06:08 AM | claude-sonnet-4-6 | 48.06 tok/s | 1.58s |
| Jun 27, 04:01 AM | gpt-image-2 | 50.99 tok/s | 3.62s |
| Jun 26, 11:33 AM | gpt-5.5 | 50.36 tok/s | 1.85s |
| Jun 25, 05:08 AM | gpt-5.5 | 56.25 tok/s | 3.34s |
| Jun 11, 10:03 AM | claude-opus-4-8 | 46.44 tok/s | 2.54s |
| May 23, 05:05 AM | glm-5.1 | 64.44 tok/s | 7.36s |
相似 API 替代服务商对比
用 309 个 LMSpeed 信号，把 Tokeness.io 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 |
|---|---|---|---|---|---|---|
| Tokeness.io tokeness-cn Tokeness.io is an OpenAI-compatible API relay hub that connects developers to mainstream LLMs including GPT, Claude, and DeepSeek through a single endpoint. | 当前服务商基线 | 47 | 1 | $0.079/M | 62 tok/s | 9980% |
| 小水管 API edge-pieixan-icu Pie-Xian (小水管) provides an OpenAI-compatible API relay with multiple endpoints including api.pie-xian.com and edge nodes. |
| 495 | 2 | $0.054/M | 260 tok/s | 9910% |
| 91VIP 91vip-futureppo-top 91VIP is a non-profit API service providing access to various AI models including Codex, Claude Code, and Open Code, with specific unlimited-use groups. |
| 383 | 2 | $0.054/M | 301 tok/s | 0% |
| VSLLM vsllm-com VSLLM runs a New API-powered AI gateway on vsllm.com for aggregated model access through a single endpoint. |
| 209 | 2 | $0.056/M | 174 tok/s | 9960% |
| CatClaw API www-catclawai-top CatClaw API provides an AI model relay with access to GPT, Claude, and other large language models. |
| 129 | 2 | N/A | 83 tok/s | 9990% |
| 初叶🍂Furry API ai-chuyel-top A free, community-supported API service providing access to various AI models for developers and users. |
| 1,021 | 152 | $18.00/M | 171 tok/s | 2970% |
| 天絮 API tianxu-api Tianxu API provides an AI model relay service with multiple access points and stable connectivity. |
| 759 | 0 | $0.014/M | 93 tok/s | 9910% |
公告
default2026/8/30
近期上线：Recently launched:glm-5.3-free，hy4-preview，glm-5.3-flash(ox-alpha,x0.3) , deepseek-v4-flash-vision-exp ,glm-5.3 ，hy3
success2026/8/24
因品牌战略升级，本平台主站已迁移到新域名：Due to the brand strategy upgrade, the main site of this platform has migrated to a new domain: Tokeness.ai
warning2026/8/24
中国大陆可用Base URL：Base URL available in mainland China: https://n.tokeness.dev/v1
ongoing2026/8/16
DeepSeek 官方已实行峰谷定价，Tokeness 已同步跟随，计费规则与官方一致，价格始终为官方价的 60%（比官方便宜 40%），峰谷同享。DeepSeek has rolled out official peak/off-peak pricing and Tokeness has followed in sync — billing rules match the official API, and our price stays at 60% of the official rate (40% cheaper) for both peak and off-peak.
success2026/8/13
deepseek-v4-pro 已更新到 0813 正式版，Agent 能力大幅增强，基准测试逼近 Opus 5，价格比官方便宜40%。deepseek-v4-pro hhas been updated to the 0813 stable version, with significantly enhanced Agent capabilities. Its benchmark scores are approaching Opus 5, and it is 40% cheaper than the official price.
default2026/7/31
deepseek-v4-flash 已更新到 0731 正式版，Agent 能力大幅增强，基准测试远超 V4-Pro-Preview，价格比官方便宜40%。deepseek-v4-flash has been updated to the official 0731 version, with significantly enhanced agent capabilities and benchmark scores far surpassing V4-Pro-Preview.The price is 40% cheaper than the official one.
FAQ
<tnt l="zh">联系我们</tnt><tnt l="en">Contact us</tnt>
<tnt l="zh">有问题反馈或者咨询请通过邮箱联系[email protected]，一般一个工作日内回复。</tnt><tnt l="en">For feedback or inquiries, please contact us at [email protected]. Replies typically within one business day.</tnt>
<tnt l="zh">退款政策</tnt><tnt l="en">Refund Policy</tnt>
<tnt l="zh">充值的余额消耗或开票后不可退款，未消耗的充值余额可凭充值记录退款，请按量充值。</tnt><tnt l="en">Recharged balance is non-refundable once consumed. Unused balance can be refunded with recharge records. Please recharge according to your usage.</tnt>
<tnt l="zh">服务可用性</tnt><tnt l="en">Service Availability</tnt>
<tnt l="zh">如果出现某模型不可用，可暂时切换模型，一般是上游问题： OpenAI：https://status.openai.com/ Claude：https://status.claude.com/ Gemini：https://aistudio.google.com/status 如果上游可用性没问题，请联系本站反馈。</tnt><tnt l="en">If a model is unavailable, try switching to another model temporarily — it's usually an upstream issue: OpenAI: https://status.openai.com/ Claude: https://status.claude.com/ Gemini: https://aistudio.google.com/status If upstream services are working fine, please contact us to report the issue.</tnt>
<tnt l="zh">付款后没到账？</tnt><tnt l="en">Payment made but not credited?</tnt>
<tnt l="zh">支付成功后通常实时到账，一般情况下几分钟内也会完成到账。若您支付后立即关闭了页面，可能导致系统回调未完成，此时会触发兜底机制，在几分钟后自动完成到账。如超过 15 分钟仍未到账，请联系我们为您处理补单。</tnt><tnt l="en">Payments are usually credited in real-time, and in most cases within a few minutes. If you close the page immediately after payment, the system callback may not complete — a fallback mechanism will then be triggered and automatically credit your account within a few minutes. If the balance is still not credited after 15 minutes, please contact us to manually process the order.</tnt>
<tnt l="zh">Codex 非 GPT 模型 报404/500？</tnt><tnt l="en">Codex non-GPT models returning 404/500 errors?</tnt>
<tnt l="zh">因为Codex只支持 Responses API ，而国产模型大多只支持 Messages 和 Completions API，导致端点不匹配。请使用支持的访问端点。</tnt><tnt l="en">Codex only supports the Responses API, while most Chinese models only support the Messages and Completions APIs, causing an endpoint mismatch. Please use a supported access endpoint.</tnt>
<tnt l="zh">需对公开票？</tnt><tnt l="en">Need a corporate invoice?</tnt>
<tnt l="zh">充值>50元可对公开票，需要请联系我们。</tnt><tnt l="en">Corporate invoices available for recharges over 50 RMB. Please contact us if needed.</tnt>
注释
- 健康检查： 统计范围：72 小时图表和近期可用率只衡量 API 连通性，每个柱形汇总一小时内的检查结果。 检查目标：LMSpeed 会优先尝试已配置的健康检查 URL 和 Provider 状态 URL，再尝试从已知 API Host 与近期测速 Base URL 推导出的 API 端点；只有官网地址本身明显是 API 地址时才会作为候选。 探测步骤：每个候选地址依次进行 DNS 解析、TCP 连接、HTTPS 的 TLS 握手，以及自动跟随重定向的 HTTP HEAD 请求；找到首个可达地址后停止尝试。 可达标准：所有必需的网络步骤都必须成功。除被归类为拦截的状态外，低于 500 的 HTTP 响应均视为可达；其中 401 表示需要鉴权，但也证明 API 端点已响应。 拦截结果：HTTP 403、429、521、525、530，以及识别出的 WAF 或 Cloudflare 挑战会标记为“被拦截”，不计入可用率，因为 LMSpeed 无法据此判断 API 本身是否宕机。 模型可用性：配置专用测试 Key 后，LMSpeed 会向推导出的 /models 端点发送带鉴权的 GET 请求，并将返回的模型 ID 与该 Provider 收录的模型进行比对。逐模型结果显示在“模型与价格”中，不计入 Provider 连通性百分比。 超时限制：TCP 连接、TLS 握手、HTTP 连通性请求和模型请求分别使用 20 秒超时；尝试多个候选地址时，整轮检查可能超过 20 秒。 检查频率：后台 Worker 默认每 5 分钟检查一次所有 Provider；72 小时图表会将这些采样汇总为小时柱形，运营方也可以调整调度。 限制说明：自动抽样结果不构成 SLA，也不能保证账户额度、所有模型、所有地区或实际补全请求始终成功；作出运维决策前请同时查看 Provider 官方状态页。
- Domain Rating 数据来自 Ahrefs。它是基于外链估算的 0–100 域名强度信号，不代表 API 速度或可用性。
- 页面上的公告和 FAQ 来自该服务商的 NewAPI 状态快照（如果可用）。LMSpeed 会保存服务商状态源中的原始内容和可选英文翻译，并在页面上展示对应语言字段。