Pay up to 82% less for the models you already run.你已经在跑的模型，最高少付 82%。
278 AI models covering text, image, audio and video. One OpenAI-compatible endpoint. Change two values, keep your code, and every request is served by the cheapest healthy line first.278 个大模型，覆盖文本、图片、音频、视频模型。一个 OpenAI 兼容端点。改两个值、代码不动，每次请求先由最便宜的可用线路承接。
See the discount before you send a token.发出第一个 token 之前，先看清折扣。
Struck-through is the provider's list price. Bold is what you pay on the lowest-priced tier that serves the model. Image, audio and video models are priced per task.划线价是官方价，加粗是承接该模型的最低价档实付。图片、音频、视频模型按任务计价。
| Model模型 | Input / 1M输入 / 百万 | Output / 1M输出 / 百万 | Pricing价目 | Discount降幅 | Group组 | |||
|---|---|---|---|---|---|---|---|---|
| claude-opus-5-5Anthropic · text文本 | Save 82%省 82% | Value | ||||||
| gpt-6-astraOpenAI · text文本 | Save 82%省 82% | Value | ||||||
| gpt-6.1-solOpenAI · text文本 | Save 82%省 82% | Value | ||||||
| glm-5.3Z.ai · text文本 | Save 82%省 82% | Value | ||||||
| grok-4.7xAI · text文本 | Save 82%省 82% | Value | ||||||
| gemini-3.8-flashGoogle · text文本 | Save 82%省 82% | Value | ||||||
| gpt-5.6-lunaOpenAI · text文本 | Save 50%省 50% | Saver | ||||||
| text-embedding-3-largeOpenAI · text文本 | N/A不适用 | Save 50%省 50% | Saver | |||||
| jev-1.13.0Other · text文本 | N/A不适用 | Save 50%省 50% | Saver | |||||
| gpt-5.6-terraOpenAI · text文本 | Save 82%省 82% | Value | ||||||
| gpt-5.4-nanoOpenAI · text文本 | Save 50%省 50% | Saver | ||||||
| gpt-5.4-miniOpenAI · text文本 | Save 50%省 50% | Saver | ||||||
| claude-fable-5-1Anthropic · text文本 | Save 82%省 82% | Value | ||||||
| claude-sonnet-5-5Anthropic · text文本 | Save 82%省 82% | Value | ||||||
| glm-5.3-flashZ.ai · text文本 | Save 82%省 82% | Value | ||||||
| glm-5.2Z.ai · text文本 | Save 20%省 20% | Balance | ||||||
| claude-sonnet-5Anthropic · text文本 | Save 82%省 82% | Value | ||||||
| qwen3.8-maxAlibaba · text文本 | Save 50%省 50% | Saver | ||||||
| deepseek-v4.1-flashDeepSeek · text文本 | Save 82%省 82% | Value | ||||||
| deepseek-v4-flash-0731DeepSeek · text文本 | Save 20%省 20% | Balance | ||||||
| deepseek-v4-pro-0813DeepSeek · text文本 | Save 82%省 82% | Value | ||||||
| MiniMax-M3MiniMax · text文本 | Save 20%省 20% | Balance | ||||||
| kimi-k3Moonshot · text文本 | Save 82%省 82% | Value | ||||||
| gemini-3.7-flashGoogle · text文本 | Save 50%省 50% | Saver | ||||||
| gemini-3.1-pro-previewGoogle · text文本 | Save 82%省 82% | Value | ||||||
Change two values. Keep everything else.改两个值，其余一切照旧。
Chat, image, audio and video all use the OpenAI request and response format, and Anthropic- and Gemini-style endpoints are available for code written against those SDKs. Task models such as video return a job you poll with one call. Your SDK, retries and logging stay exactly as they are.对话、图片、音频、视频都用 OpenAI 的请求与响应格式；按 Anthropic、Gemini SDK 写的代码也有对应端点。视频这类任务型模型返回一个任务，用一次调用轮询即可。你的 SDK、重试、日志一样不用改。
# before client = OpenAI(api_key="sk-…") # after client = OpenAI( base_url="https://modapi.ai/v1", api_key="sk-modapi-…", ) r = client.chat.completions.create( model="gpt-6-astra", messages=[{"role": "user", "content": "hi"}], )
# same client, same SDK img = client.images.generate( model="gpt-image-2", prompt="a lighthouse at dusk, film grain", size="1024x1024", ) # priced per image, by size and quality
# task models return a job you poll job = client.post("/videos", body={ "model": "veo_3_1", "prompt": "slow dolly across a rainy street", "seconds": 8, }) status = client.get(f"/videos/{job['id']}") # priced per video, by duration and resolution
Not just chat. 278 models across all modalities.不止对话。278 个模型，覆盖全模态。
One key, one bill and one log across text generation, image generation and editing, speech and music, and video.
Task models share one job format, so you poll one way for all of them.一把密钥、一张账单、一份日志，覆盖文本生成、图片生成与编辑、语音与音乐、视频。
任务型模型共用一种任务格式，轮询方式只需一种。
Text & reasoning文本与推理
GPT, Claude, Gemini, Grok, DeepSeek, Qwen, GLM, Kimi, MiniMax and embeddings.GPT、Claude、Gemini、Grok、DeepSeek、Qwen、GLM、Kimi、MiniMax 与向量模型。
36Image图片
Generation, editing and Midjourney workflows, priced by size and quality.生成、编辑与 Midjourney 流程，按尺寸与质量计价。
20Audio音频
Text-to-speech, transcription and music, priced by characters or seconds.语音合成、转写与音乐，按字数或秒数计价。
33Video视频
Veo, Grok, Kling, Vidu and more, with reference images and configurable duration.Veo、Grok、Kling、Vidu 等，支持参考图与可配置时长。
Four channel groups. Four price tiers.四个渠道组，四种价格
Our algorithms organize a large pool of idle capacity into four tiers. Lock a channel group for a fixed price.
Or use auto routing: lowest price first, with higher-priority lines as fallback.通过特定算法将大量闲置资源分为了四个档次，可锁定某一组拿固定价格。
可auto 路由：低价优先，高优线路兜底。
Value
Lowest special-price group; opt in per key.最低特价组；按密钥自选。
Saver
Exceptional value.超高性价比。
Balance
Price and reliability in balance. Recommended for most production traffic.价格与稳定兼顾，多数生产流量的推荐项。
Prime
Higher-priority lines, selected for stability. For latency- and reliability-critical workloads.高优线路，为稳定而选。延迟与可靠性敏感的负载。
Image, video, audio and music. Same key, same bill.图片、视频、音频、音乐。同一把密钥，同一张账单。
One key connects: OpenAI, Google, Midjourney, Kling, Vidu, ByteDance, xAI, MiniMax, Alibaba, PixVerse and Suno.
Each is priced per task by size, duration and quality, with the exact charge shown on every job. Long renders run asynchronously: submit, poll, collect.一个 key 全接：OpenAI、Google、Midjourney、可灵、Vidu、字节、xAI、MiniMax、阿里、PixVerse、Suno。
按任务计价，价格由尺寸、秒数、清晰度决定，每个任务都能看到实际扣费。长渲染走异步任务：提交、轮询、取回。
OpenAIGPT Image
GoogleVeo · Imagen · Gemini Image
MidjourneyV7 · Niji
KlingKling 3 video
ViduVidu Q3
ByteDanceSeedance · Seedream
xAIGrok Imagine
MiniMaxHailuo · Speech
AlibabaWan · Qwen-Image
PixVersePixVerse video
SunoMusic generationVolume gets its own price.量大，价格单谈。
- Negotiated rates per model or per tier, committed or usage-based.按模型或按档位议价，可承诺量也可按量。
- A dedicated tier and channels for your traffic, in your priority order.为你的流量单开档位与渠道，优先级由你定。
- Models you need that we don't list yet, onboarded on request.需要但我们尚未上架的模型，按需接入。
- Invoices, usage exports and per-team keys.发票、用量导出、按团队分发密钥。
We reply within one business day.一个工作日内回复。
Lower the cost of your next request.让下一次请求更便宜。
Create a key, change two values, and watch the tier column in your log.建一把密钥，改两个值，看日志里的档位列。
Get API key获取密钥