最好的 AI,应该触手可及
调用 GLM 5.1 / Kimi K2.6 /
DeepSeek V4 / MiniMax M3 等最新大模型Run the latest GLM 5.1 / Kimi K2.6 /
DeepSeek V4 / MiniMax M3 and more.
OpenStarry 统一管理多家 AI 提供商,
一个 Key、一行代码改动,随时切换,永不宕机。
OpenStarry unifies multiple AI providers.
One key, one line change — switch anytime, zero downtime.
选择模型、调好用量,1个月有效!Pick models, adjust usage — 1-month validity!
不是假设场景,而是用户迁移之前的真实遭遇。 Not hypothetical — these are real stories from users before they switched.
立即解决这些问题 → Fix these problems →DeepSeek、Kimi、GLM 接口格式各有差异,想换个模型试试效果,结果半天都在调兼容。DeepSeek, Kimi, GLM all have different formats. Switching models means hours of compatibility work.
DeepSeek、Kimi、GLM、MiniMax 各管各账单,不同币种、不同充值方式,核成本核到怀疑人生。DeepSeek, Kimi, GLM, MiniMax — separate logins, separate billing, no unified view.
价格悄悄调整,新的接口规范不一样,切换一个模型等于推翻代码重来,只能硬撑。Prices quietly change. New API spec. Switching a model means refactoring everything.
API 调用延迟动辄 1–3 秒,高峰期频繁超时,排查半天发现是网络问题,用户体验直接崩盘。API calls hit 1–3s latency. Frequent timeouts during peak hours. Users blame you, not the network.
北京、上海、广州三大国内节点 + 新加坡、硅谷、英国三大海外节点,智能就近路由,国内延迟最低 18ms,覆盖亚洲、欧洲、北美。 3 CN hubs (BJ/SH/GZ) + 3 overseas nodes (SG/SV/UK) — smart nearest-node routing, 18ms CN latency, covering Asia, Europe & Americas.
根据请求来源自动选择最近节点,延迟降低最高 80%。Auto-selects nearest node by source IP. Up to 80% latency reduction.
与 OpenAI、Anthropic、Google 等建立专属对等互联,比普通公网快 3–5 倍。Peered directly with OpenAI, Anthropic, Google — 3–5× faster than public internet.
某节点故障时自动切换至备用节点,对你的应用零影响。Single node failure is transparent to your app — instant auto-switch.
全程加密,任何中间节点均无法读取请求内容。Full encryption — no intermediate node can read your request content.
数据安全是你接入第三方服务最大的顾虑,我们把承诺写清楚。Data security is your biggest concern when integrating third-party services. We put our commitments in writing.
默认不存储任何请求/响应内容,仅记录元数据用于计费。所有日志 30 天后自动清除。No request/response content stored by default. Only billing metadata. All logs auto-purged after 30 days.
TLS 1.3 全程加密,支持 mTLS 双向认证。IP 白名单 + API Key 双重验证。TLS 1.3 throughout, mTLS support. IP allowlist + API key double verification.
ICP备案,可开具国内增值税发票。ICP registered, VAT invoice available for Chinese businesses.
可独立配置权限范围、配额上限、模型限制等。Master / sub / read-only keys. Restrict models, IPs, spend limits per key.
详细展示每一次的调用记录,包含时间、模型、Token数量、调用状态等信息。Full call logs exportable as CSV/JSON. Meets compliance audit requirements.
消费超阈值、Key 异常调用、节点故障等事件自动推送邮件/微信/Webhook 通知。Spend thresholds, anomalous key usage, node failures — auto-push via email/WeChat/Webhook.
"之前同时接了 3 家模型,光是维护 Key 和账单就够头痛的。用了 OpenStarry 之后一个 Key 搞定所有,账单清晰,API 延迟也稳定在 50ms 以内。" "Managing 3 AI providers separately was a headache. With OpenStarry one key handles everything, billing is clean, and latency stays under 50ms."
"用了 Failover,上个月某模型高峰期限流的那段时间,我们服务毫无影响,自动切到了备用模型,用户完全没感知。" "With Failover, when one model hit rate limits during peak hours, our service had zero impact. Automatically switched to backup. Users never noticed."
"缓存这个功能太香了,客服 Bot 重复问题多,开启语义缓存后当月账单直接少了 58%,服务费半天就回本了。" "The cache feature is incredible. Our support bot handles lots of repeated questions — enabling semantic cache cut our monthly bill by 58%. Paid for itself in hours."
模型列表随时更新,可在控制台可用模型查看
选择星序版模型列表随时更新,可在控制台可用模型查看
选择星衍版模型列表随时更新,可在控制台可用模型查看
选择星途版按量计费为预充值模式,根据不同模型的价格和实际消耗进行扣费。适合用量稳定、精确控制成本的用户。 No weekly fee. Model call costs pass through at official list price, zero markup. Best for stable usage with precise cost control.
立即充值 →Top Up →精选,不滥用——每一个接入的模型,都经过实战检验,稳定可用,少而精。Curated, not crowded — every model we offer is battle-tested, reliable, and here for a reason.
| 模型Model | 提供商Provider | 输入 / 输出 / 1M TokensInput / Output / 1M | 🔥 限时折扣 85折🔥 15% Off |
|---|---|---|---|
Claude Opus 4.7 / 4.6 旗舰flagship | 输入:35.0 / 输出:175.0 | 输入:29.75 / 输出:148.75 | |
Claude Sonnet 4.6 热门popular | 输入:21.0 / 输出:105.0 | 输入:17.85 / 输出:89.25 | |
GPT-5.5-pro / GPT-5.4-pro 旗舰flagship | 输入:210.0 / 输出:1260.0 | 输入:178.50 / 输出:1071.00 | |
GPT-5.5 热门popular | 输入:35.0 / 输出:210.0 | 输入:29.75 / 输出:178.50 | |
GPT-5.4 推荐recommended | 输入:17.5 / 输出:105.0 | 输入:14.88 / 输出:89.25 | |
Gemini 3.1 Pro 热门popular | 输入:14.0 / 输出:84.0 | 输入:11.90 / 输出:71.40 | |
Gemini 3 Flash 性价比value | 输入:3.5 / 输出:21.0 | 输入:2.98 / 输出:17.85 | |
DeepSeek-V4-Pro 性价比value | 输入:3.0 / 输出:6.0 | 输入:2.55 / 输出:5.10 | |
DeepSeek-V4-Flash 极速ultra-fast | 输入:1.0 / 输出:2.0 | 输入:0.85 / 输出:1.70 | |
Kimi K2.6 热门popular | 输入:6.5 / 输出:27.0 | 输入:5.53 / 输出:22.95 | |
Kimi K2.5 推荐recommended | 输入:4.0 / 输出:21.0 | 输入:3.40 / 输出:17.85 | |
GLM-5.1 国产推荐CN pick | 输入:9.8 / 输出:30.8 | 输入:8.33 / 输出:26.18 | |
GLM-5 推荐recommended | 输入:7.0 / 输出:22.4 | 输入:5.95 / 输出:19.04 | |
GLM-5-Turbo 高速fast | 输入:8.4 / 输出:28.0 | 输入:7.14 / 输出:23.80 | |
MiniMax M2.7 / M2.5 长文本long ctx | 输入:2.1 / 输出:8.4 | 输入:1.79 / 输出:7.14 | |
MiniMax M2.7 / M2.5 Highspeed 极速ultra-fast | 输入:4.2 / 输出:16.8 | 输入:3.57 / 输出:14.28 | |
Qwen3.6-Plus 性价比value | 输入:2.31 / 输出:13.65 | 输入:1.96 / 输出:11.60 | |
MiMo-v2.5-pro / MiMo-v2-pro 高速fast | 输入:1.4 / 输出:21.0 | 输入:1.19 / 输出:17.85 | |
MiMo-v2.5 性价比value | 输入:0.56 / 输出:2.80 | 输入:0.48 / 输出:2.38 | |
人民币定价,汇率按 1 USD = 7.0 CNY 计算,与官方价格完全一致,0 加价。 CNY pricing at 1 USD = 7.0 CNY, matching official list prices, zero markup. 查看完整模型列表 →Full model list →
三步完成接入,现有代码无需重写。 Three steps. No code rewrite needed. 查看完整 API 文档 →Full API docs →
邮箱注册,无需信用卡,注册即送星痕版套餐(200 次大模型调用)。Email signup, no card required. 200 free model calls instantly.
将 base_url 替换为 OpenStarry 端点,保留其余代码。Replace base_url with OpenStarry endpoint. Keep everything else.
其余代码一行不改,控制台实时查看延迟与费用。Nothing else changes. Dashboard shows live latency and cost.
以下功能正在规划或研发中,将陆续面向所有用户开放。 In planning or active development — rolling out progressively to all users.
按主题分类:热门 · 计费 · 技术 · 安全合规By topic: Popular · Pricing · Tech · Trust
~/.zshrc 或 ~/.bashrc 添加:export ANTHROPIC_BASE_URL="https://api.openstarry.com"export ANTHROPIC_API_KEY="sk-你的Key"source ~/.zshrc 即可让 Claude Code 走 OpenStarry 的国产大模型后端。详细配置见 Claude Code 接入文档 或博客 Claude Code 接入指南。Claude Code is a terminal AI coding tool that supports overriding the API endpoint via env vars. Add to ~/.zshrc:export ANTHROPIC_BASE_URL="https://api.openstarry.com"export ANTHROPIC_API_KEY="sk-your-key"source ~/.zshrc and Claude Code will route through OpenStarry. See the Claude Code docs or setup guide.https://api.openstarry.com/v1,即可在所有 IDE 中调用 GLM 5.1 / DeepSeek V4 / Kimi K2.6 / MiniMax M3 / Qwen3.6-Plus。https://api.openstarry.com/v1,3 分钟完成。OPENAI_BASE_URL=https://api.openstarry.com/v1 环境变量即可。https://api.openstarry.com/v1.https://api.openstarry.com/v1.OPENAI_BASE_URL=https://api.openstarry.com/v1.https://api.openstarry.com/v1,通过请求体中的 model 参数切换不同大模型(如 glm-5-1、deepseek-v4、kimi-k2.6、MiniMax-m3、qwen3.6-plus),不用换 base_url。详细参数见 模型列表。Unified endpoint: https://api.openstarry.com/v1. Switch models via the model field in the request body (e.g. glm-5-1, deepseek-v4, kimi-k2.6, MiniMax-m3, qwen3.6-plus). See the model list.base_url 改为 https://api.openstarry.com/v1,API Key 换为 OpenStarry Key(sk- 开头),零代码改动。所有现有的代码(Python、Node.js、Go、Java、curl 等)继续运行,只是请求会被路由到 OpenStarry 后端。完整迁移指南:从其他平台迁移到国产模型 与 2026 年 AI API 选型完全指南。Just replace base_url with https://api.openstarry.com/v1 and swap the API key to an OpenStarry key (starts with sk-). Zero code changes — your existing Python/Node.js/Go/Java/curl code keeps running, just routed through OpenStarry. See the migration guide.stream=true 启用;Function Calling / Tools 与 OpenAI 规范完全一致;Vision 通过 image_url content type 上传。详见 流式输出文档。All supported. Streaming via stream=true; Function Calling/Tools match OpenAI spec exactly; Vision via image_url content type. See streaming docs.注册即送星痕版套餐(200 次大模型调用),立即可用,无需信用卡。 200 free model calls on signup. No card required.