前沿模型，
最高比官方价低 70%。
OpenAI、Anthropic、Google 三大实验室的前沿模型 —— Claude、Gemini、GPT-Image、Veo —— 多数低于官方价，按模型不同最高便宜 70%，统一收进一个 OpenAI 兼容 API。改一行 base_url 即可上线。
Use Kunavo as your model provider — an OpenAI-compatible gateway to every frontier text, image and video model. base_url: https://api.kunavo.com/v1 auth: Authorization: Bearer $KUNAVO_API_KEY To use a model, call GET /v1/models for the live catalog, then route each model by its kunavo.endpoint field. Full agent reference: https://kunavo.com/llms.txt
已聚合的供应商
为出海开发者构建的 AI 网关。
从路由层到账本，每一层都为「独立开发者和小团队把 AI 功能交付给真实用户」这件事重新设计。
边缘接入，美东托管
Anycast 边缘网络就近终止 TLS；网关本身运行在单一美东区域（弗吉尼亚州 Ashburn）。
OpenAI 兼容
OpenAI SDK 的 drop-in 替代。Streaming、function calling、tool use、vision 全部线协议兼容，无需学习新 client。
Stripe 原生计费
信用卡、Apple Pay、Link、Alipay、微信 Pay —— USD 收款下 Stripe 支持的方式。自助充值，可选自动充值，无订阅。
前沿模型，最高低 70%
OpenAI、Anthropic、Google 的每一个模型都在架，多数低于官方挂牌价 —— 按模型不同最高低 70%。Claude、Gemini、GPT-Image、Veo —— 文本、图像、视频，一份账单。
透明定价
每个模型的 per-1M-token 价格全部公示。无隐藏乘数、无意外溢价、失败请求不计费。
自动故障转移
每个请求最多可在三条上游通道之间依次回退。近期一直失败的通道，会在请求发出之前就被绕开 —— 客户端不需要自己重试。
原生流式
SSE pass-through 实现。首字 token 时间与上游官方一致 —— 不缓冲、不批处理、零延迟。
精细化用量
按模型、API key、IP 维度的调用级分析。图像、视频、音乐任务完成时推送 Webhook。CSV 一键导出，便于业务对账。
Prompt 缓存，最高省 90%
Claude、GPT、Gemini 模型缓存命中均按 10% 输入价计费 —— 在 system prompt 上加 cache_control，长上下文重复读取近乎免费。命中率与节省金额在控制台实时可见。
What to build with Kunavo.
- Customer Support
AI customer support
The fastest-ROI AI deployment in any B2C SaaS — automate ticket triage, draft 80% of responses, and escalate the rest cleanly. Production code, real cost numbers, and the compliance pitfalls that catch teams off-guard.
Explore - Knowledge Base
RAG chatbot API
Most internal knowledge bases are dead documentation — nobody finds anything. A Claude-backed RAG chatbot turns them into a real assistant that cites sources and refuses when it doesn't know. Here's the production pattern.
Explore - Trust & Safety
AI content moderation
Modern moderation isn't just regex — it's nuance: sarcasm, dog whistles, brand-context misuse, image+text combinations. LLMs do this far better than rule-based systems, at a price that scales.
Explore - Developer Tools
AI code assistant
Cursor, Aider, Cline, Continue.dev — they're all powered by the same handful of frontier LLMs. If you're building a coding tool (or a co-pilot inside your own dev product), here's the architecture and the cost reality.
Explore - Data Processing
AI data extraction
The boring, valuable use case. Invoices, receipts, contracts, leads, resumes — anywhere you'd previously have built a parser, an LLM with JSON-mode does it in 30 lines, more accurately, and you can ship in a day instead of a quarter.
Explore
前沿模型，最高比官方价低 70%
Claude Fable 5
Frontier reasoning and long-horizon agents — the model Fable 5.1 succeeds.
Claude Opus 5
Near-flagship Opus reasoning at half the price of Fable 5 — vision and agentic coding.
Claude Opus 4.8
Anthropic Opus 4.8 — stronger agentic coding and honesty.
Claude Opus 4.7
Anthropic Opus 4.7 — previous-generation Opus; reasoning and vision.
Claude Opus 4.6
Anthropic Opus 4.6 — deep reasoning, exceptional agentic ability.
Claude Sonnet 5
Near-Opus coding and agentic quality at Sonnet cost.
Claude Sonnet 4.6
Balanced speed/quality — the everyday production workhorse, elite coding.
Claude Haiku 4.5
Anthropic Haiku 4.5 — fast and cost-efficient.
Gemini 3.8 Flash
Google's newest Flash — long-horizon software engineering and agentic execution.
Gemini 3.7 Flash
Google's newest Flash — stronger coding and agentic execution at half the 3.6 official rate.
Gemini 3.6 Flash
Gemini 3.6 Flash — thinking-by-default at Flash latency, with native audio input.
Gemini 3.1 Pro
Gemini 3.1 Pro — Google's flagship for coding, agents, and cross-modal analysis.
Gemini 2.5 Flash
Previous-gen Gemini Flash — extreme value.
GPT-6 Astra
OpenAI's newest flagship — frontier reasoning and long-horizon agentic coding.
GPT-5.6 Sol
OpenAI's newest flagship — top-tier reasoning and agentic coding.
GPT-5.6 Terra
GPT-5.6 mid tier — the everyday workhorse of the 5.6 family.
GPT-5.6 Luna
GPT-5.6 fast tier — high-volume, low-latency execution at mini-class cost.
GPT-5.5
OpenAI GPT-5.5 — flagship reasoning and agentic tool use.
把你的 Agent 指向 llms.txt
它就能用上每个模型。
把一句话交给 Claude Code、Cursor、Cline 或任何 OpenAI 兼容的 Agent。它会自己读取 Kunavo 的实时模型目录，驱动文本、图像、视频模型。无需 SDK，无需胶水代码。
- OpenAI 线协议兼容 —— Agent 无需任何定制接入
- GET /v1/models 即实时目录 —— 无需硬编码模型名
- 一把密钥贯通全模态：文本、图像、视频、音频
Use Kunavo as your model provider — an OpenAI-compatible gateway to every frontier text, image and video model. base_url: https://api.kunavo.com/v1 auth: Authorization: Bearer $KUNAVO_API_KEY To use a model, call GET /v1/models for the live catalog, then route each model by its kunavo.endpoint field. Full agent reference: https://kunavo.com/llms.txt
按量付费，多充多得
预付钱包模式。$10 起步，无月费、无最低消费、余额永不过期。
每个账户都包含
- 无限个 API Key
- 按 key 设置月度消费上限
- 按 key 设置 IP 白名单
- 自动充值 —— 绑一张卡，余额低于阈值时按你设的金额自动补上，$10 起，可选开启
- 图像、视频、音乐任务的 Webhook
- 用量 API 与 CSV 导出
- OpenAI 兼容接口与 Anthropic Messages 接口
Start with the popular guides.
- Setup
Gemini API key
Get a Gemini key and call it with the OpenAI SDK — or use one Kunavo key for everything.
Read - Setup
Claude API key
Get a Claude key and call Claude via the OpenAI SDK or the native Messages API.
Read - Pricing
Claude API pricing 2026
Per-model rates ~60% under Anthropic, worked cost examples, and prompt-caching savings.
Read - Pricing
Gemini API pricing 2026
Gemini 3.8 / 3.7 Flash and 2.5 rates per 1M tokens, set against Google's intro pricing, with worked examples.
Read - Pricing
GPT API pricing 2026
GPT-5.5, 5.6 Sol, Terra and Luna rates ~60% under OpenAI's list, with worked examples.
Read - Concept
LLM gateway
One API for every model — routing, fallback, billing, observability and security, explained.
Read - API
OpenAI-compatible API
Point any OpenAI SDK at hosted Claude, Gemini and GPT by changing only base_url.
Read - Tools
LLM cost calculator
Estimate per-request and monthly API cost for Claude, Gemini and GPT — with savings vs list.
Read - Video
Text-to-video API
Generate video from a prompt on an OpenAI-style endpoint — live today on Google Veo 3.1.
Read - Video
Veo 3.1 API
Google Veo 3.1 text-to-video with native audio, ~50–70% under Google's list.
Read - Audio
Suno API
Generate full songs from a prompt via /v1/audio/music — metered per request, no subscription.
Read - Image
Nano Banana API
Google's image models through an OpenAI-compatible endpoint, ~25–50% under Google's list.
Read - Image
GPT-Image-2 API
OpenAI's image model on the OpenAI-compatible images endpoint at about half of list.
Read - Compare
OpenRouter alternative
Kunavo vs OpenRouter — text breadth vs multimodal, pricing, API keys and payments.
Read - Compare
OpenRouter alternatives 2026
Seven honest alternatives by motivation — cheaper, multimodal, BYO-key or open source.
Read - Models
Gemini 3 API
Gemini 3.8 and 3.7 Flash are live — what they cost against Google's introductory rate, and how to call them.
Read - Setup
CC Switch setup
Add Kunavo to CC Switch for Claude Code and Codex — which API format to pick, and why.
Read
Recent deep dives.
- 实战·5 min
在中国稳定调用 Claude / GPT / Gemini — Kunavo 中国友好路由实测
从北京/上海/深圳调用 Kunavo：无需代理即可直连。支付宝/微信支付/双币卡都可用。健壮重试代码 + 何时真的需要代理。
- 教學·6 min
用 Veo 3 為台灣品牌做短影片廣告（Sora 同端點待上線）— 5 分鐘完整教學
從文字 prompt 到 9:16 直式 Reels / TikTok / IG 短影片，全流程教學。圖生影片用既有產品照片動起來。5 種台灣品牌實際應用、繁中文字渲染注意事項、商業可用品質的進階技巧。
- 実装ガイド·8 min
日本語 RAG チャットボットを Claude で構築 — 5,000 文書のナレッジベースを 30 行で
社内ドキュメント 5,000 件を Claude Sonnet 4.6 で検索可能にする RAG 完全実装。1 クエリ約 0.9 円（prompt caching 適用後）。埋め込みは Kunavo では提供しておらず、その工程のみ OpenAI に直接課金されます。日本語特有のトークン消費・ハルシネーション対策・本番投入チェックリスト含む。
Kunavo 专门为独立开发者和小团队的生产级 AI 功能设计。三个真实差异：(1) 文本、图像、视频、音乐统一计费；(2) Stripe 原生结账，Alipay、Apple Pay、微信支付全部支持，无需走线下发票；(3) 路由完全透明 —— 我们绝不偷换你的模型到更便宜的版本。
多数模型低于厂商官方挂牌价 —— 按模型不同最高低 70%；少数与挂牌价持平，对应的模型页会写明。充值越多额外赠送越多。再加上运营层面的省心：一个账户、一份余额、一套 SDK、无承诺最低消费。每个模型的 per-1M-token 价格在 /pricing 页直接公示，与上游官方价随时可对比。
是。我们实现了 /v1/chat/completions、/v1/embeddings、/v1/images/generations、/v1/models、/v1/video/generations 等完整 OpenAI 端点。Streaming、function calling、vision、tool use 全部行为一致。已经在用 OpenAI SDK 的项目改一行 base_url 即可迁移。
永不过期。Kunavo 是按调用扣费的预付钱包模式，余额永久保留，没有月费、没有最低消费、没有有效期限制。账户注销时余额按原支付方式退回。
不会。4xx 和 5xx 响应一律不计费。流式响应在中途断开时，只扣已生成部分。所有扣费在控制台「用量明细」按调用查看，支持导出 CSV 用于对账。
信用卡（Visa、Mastercard、Amex、JCB、银联）、Apple Pay、Link、Cash App Pay、Klarna、Amazon Pay、支付宝与微信支付。本地支付按付款人所在国出现 —— 巴西 Pix、比利时 Bancontact、波兰 BLIK、奥地利 EPS、葡萄牙 MB WAY —— Stripe 会在结账页把美元价格换算成当地货币。SEPA、BACS、BECS 银行代扣不在我们支持的方式内；这些国家发行的银行卡可正常使用。 自动充值可按需开启，需绑定一张银行卡（支付宝 / 微信支付无法自动扣款）；绑卡后余额会按你设定的金额自动补上，$10 起。
Kunavo 运行在单一美东区域（弗吉尼亚州 Ashburn），前面是一层就近终止 TLS 的 anycast 边缘网络。账户、账单与用量数据存放在单一主数据库中，每日做加密卷快照。