API REFERENCE

型号

列出 API 密钥上可用的每个模型，包括功能标志、上下文限制和每个令牌的定价。

API 其余部分所依赖的一切都可通过一次调用获取：哪些模型接受 images、tools、reasoning 或 caching，它们的 context window，以及每 token 的价格。请查询 models endpoint 并按下面的能力标志进行筛选，而不要在应用中硬编码模型列表。

请注意，created 字段是该列表的当前服务器时间，而非各模型的发布日期；status 字段反映实时健康状况 —— 请重新拉取它，而不要长时间缓存。

GET /v1/models

返回您现在可以使用的模型列表。经过身份验证的请求会包含您在仪表板中启用的所有按用户配置的模型。其结构与 OpenAI 的 GET /v1/models 并附带额外的 Airforce 能力字段。

GEThttps://api.airforce/v1/models

例子

curl https://api.airforce/v1/models \
  -H "Authorization: Bearer sk-air-YOUR_API_KEY"

响应形状

顶层信封是 { object: "list", data: Model[] }. 每个 Model 具有以下字段 - 标记为可选的字段在未知时可以省略。

Parameter	Type	Required	Description
id	string	Optional	Public model identifier you pass as the "model" field on every other endpoint, e.g. "claude-sonnet-4.6" or "gpt-5.1-chat".
object	string	Optional	Always "model".
created	integer	Optional	Unix timestamp at which this listing was generated (current server time) — not a stable per-model creation date.
owned_by	string	Optional	Vendor name: "openai", "anthropic", "google", "deepseek", …
catalog_id	string	Optional	Canonical vendor identifier for the underlying model family (e.g. "anthropic/claude-sonnet-4-6"). Stable across Airforce ID changes.
context_length	integer	Optional	Total context window in tokens (input + output).
max_output_tokens	integer	Optional	Hard cap on a single response. Use this as the upper bound when picking max_tokens.
input_modalities	array	Optional	Subset of ["text", "image", "audio", "document", "video"] the model accepts as input.
output_modalities	array	Optional	Subset of ["text", "image", "audio", "video"] the model can produce.
supports_chat	boolean	Optional	Available on /v1/chat/completions and /v1/messages.
supports_streaming	boolean	Optional	Honors stream: true.
supports_vision	boolean	Optional	Accepts image_url content blocks.
supports_tools	boolean	Optional	Accepts the tools / tool_choice parameters.
supports_reasoning	boolean	Optional	Honors reasoning_effort / thinking / thinking_budget.
supports_documents	boolean	Optional	Accepts document content blocks (PDF, etc.).
supports_caching	boolean	Optional	Honors cache_control on Anthropic-style system prompts.
supports_audio_input	boolean	Optional	Accepts audio content blocks (multimodal STT-in-chat).
supports_video_input	boolean	Optional	Accepts video content blocks.
knowledge_cutoff	string	Optional	Vendor-published knowledge cutoff month, e.g. "2025-09".
status	string	Optional	Live health: "operational" \| "degraded" \| "partial_outage" \| "major_outage" (legacy "stable" / "down" / "offline" may also appear).
pricepermilliontokens	integer	Optional	Input cost in cents per 1M tokens. Output side has output_pricepermilliontokens.
cache_read_pricepermilliontokens	integer	Optional	Discounted rate for cache-hit input tokens (Anthropic-style caching).

响应示例

{
  "object": "list",
  "data": [
    {
      "id": "claude-sonnet-4.6",
      "object": "model",
      "created": 1740000000,
      "owned_by": "anthropic",
      "catalog_id": "anthropic/claude-sonnet-4-6",
      "context_length": 200000,
      "max_output_tokens": 64000,
      "input_modalities": ["text", "image", "document"],
      "output_modalities": ["text"],
      "supports_chat": true,
      "supports_streaming": true,
      "supports_vision": true,
      "supports_tools": true,
      "supports_reasoning": true,
      "supports_documents": true,
      "supports_caching": true,
      "knowledge_cutoff": "2025-08",
      "status": "operational",
      "pricepermilliontokens": 300,
      "output_pricepermilliontokens": 1500,
      "cache_read_pricepermilliontokens": 30
    }
    // … more models
  ]
}

按能力过滤

能力筛选在客户端使用下面的标志完成。最常见的筛选：

const res = await fetch('https://api.airforce/v1/models', {
  headers: { Authorization: `Bearer ${API_KEY}` },
});
const { data } = await res.json();

// Vision-capable chat models
const visionModels = data.filter(m => m.supports_chat && m.supports_vision);

// Cheapest reasoning model with at least 100k context
const reasoning = data
  .filter(m => m.supports_reasoning && (m.context_length ?? 0) >= 100_000)
  .sort((a, b) => (a.pricepermilliontokens ?? 0) - (b.pricepermilliontokens ?? 0))[0];

// All models from one vendor, by canonical catalog family
const allClaude = data.filter(m => m.catalog_id?.startsWith('anthropic/'));

存在一个可选的服务端查询参数：GET /v1/models?channels=1 会额外返回各 channel 的别名条目（供具备路由感知能力的客户端使用）；不带该参数时这些别名会被省略。

定价字段

全部 *pricepermilliontokens 字段是每 1 000 000 个令牌的整数美分数。除以 100 即可得到每百万令牌的美元价格。

Parameter	Type	Required	Description
pricepermilliontokens	integer	Optional	Input tokens, cents per 1M.
output_pricepermilliontokens	integer	Optional	Output tokens, cents per 1M.
cache_write_5m_pricepermilliontokens	integer	Optional	Cache-write rate for 5-minute TTL caches (Anthropic).
cache_write_1h_pricepermilliontokens	integer	Optional	Cache-write rate for 1-hour TTL caches (Anthropic).
cache_read_pricepermilliontokens	integer	Optional	Cache-read rate (typically 10× cheaper than uncached input).

模型状态和弃用

Parameter	Type	Required	Description
operational	status	Optional	Healthy — recent calls are succeeding. The usual value for an available model.
degraded	status	Optional	Callable, but with elevated errors or latency.
partial_outage	status	Optional	A subset of upstream channels is failing.
major_outage	status	Optional	Most or all calls are currently failing.
stable / down / offline	status	Optional	Legacy or admin-set values that may also appear.

硬性禁用的模型并不是一个状态值——它们会被完全排除在 /v1/models 之外（对每个调用方均如此，包括管理员），且只能通过管理工具以精确名称调用。

想要可浏览的视图吗？

这 /模型页面通过分组、搜索和定价比较来呈现相同的数据。