NezhaGate API Developer Docs

Quick Start

Just swap in the Base URL and API Key, and your existing OpenAI SDK works out of the box.

https://nezhagate.com/v1

curl https://nezhagate.com/v1/chat/completions -H "Authorization: Bearer YOUR_API_KEY" -H "Content-Type: application/json" -d '{"model": "gpt-5.5", "messages": [{"role": "user", "content": "Write a short product intro for an AI API"}], "stream": false}'

Full request parameters, examples and async usage for each model live in that model's own doc — open it from the Models & pricing table below.

More runnable examples (Python / Node / curl, including the async image and video flow): github.com/gaoorange/nezhagate-api-examples

Authentication

All API calls require a Bearer Token.

Authorization: Bearer YOUR_API_KEY
Content-Type: application/json

Models & Pricing

Full request parameters, examples and async usage for each model live in that model's own doc — open it from the Models & pricing table below.

Use it in an app (no code)

In any app that speaks the OpenAI API, fill in these three fields — no programming needed:

Verified with ChatBox, Cherry Studio, NextChat, LobeChat, Immersive Translate, Open WebUI, etc.

Step-by-step client guides: SillyTavern · Cherry Studio · Cline

Native Anthropic API (Claude Code)

Beyond the OpenAI-compatible route, Claude models also expose the native Anthropic Messages API. Point Claude Code (or any Anthropic SDK) at the base URL below and use your gateway key as x-api-key — thinking, tool use and prompt caching are all preserved natively.

export ANTHROPIC_BASE_URL=https://nezhagate.com/anthropic
export ANTHROPIC_API_KEY=YOUR_API_KEY
# claude-sonnet-4-6 / claude-opus-5 / claude-fable-5 / claude-opus-5-5 / claude-sonnet-5-5

curl https://nezhagate.com/anthropic/v1/messages -H "x-api-key: YOUR_API_KEY" -H "anthropic-version: 2023-06-01" -H "Content-Type: application/json" -d '{"model": "claude-sonnet-4-6", "max_tokens": 1024, "messages": [{"role": "user", "content": "Hello"}]}'

Models: claude-opus-5-5, claude-sonnet-5-5, claude-opus-5, claude-sonnet-5, claude-fable-5, claude-sonnet-4-6. Billed exactly like /v1, per token.

Prompt-cache billing: cache hits (cache read) are billed at 1/10 (0.1×) of the input rate, cache writes at 1.25× (per-model rates on the pricing page). A hit needs a stable, long-enough prefix (Claude Opus ≥ 4096 tokens, Sonnet ≥ 2048) repeated within a short window — Claude Code with a large CLAUDE.md / system prompt over a real codebase hits every turn, cutting input cost to ~1/10. Short Playground chats are too small to cache.

Video Generation (async)

Text-to-video / image-to-video are async jobs: the call returns HTTP 202 + a job id immediately ({"id":"img_...","object":"video.generation.job","status":"queued"}); poll GET /v1/videos/jobs/{id} until status=succeeded, result in data[0].url (a re-hosted mp4). 1-12 min per clip. Two billing modes: Veo 3.1 is billed per clip; Seedance 2.x and MiniMax H3 are billed per second (rate x duration). Both hold exactly what they settle and refund in full on failure. Set the aspect ratio via size (16:9 landscape is the default); for image-to-video pass a first-frame reference in image.

curl https://nezhagate.com/v1/videos/generations -H 'Authorization: Bearer YOUR_API_KEY' -H 'Content-Type: application/json' -d '{"model": "veo-3.1", "prompt": "a cat surfing at sunset", "tier": "fast", "resolution": "1080p", "size": "16:9"}'

curl https://nezhagate.com/v1/videos/jobs/img_3f9a...c2 -H 'Authorization: Bearer YOUR_API_KEY'

Video models: veo-3.1 gemini-omni-flash kling-4.0 seedance-2.5-token seedance-2.5-token-draft seedance-2.5 seedance-2.0 seedance-2.0-fast seedance-2.0-clip seedance-2.5-30s wan3.0-video wan3.0-video-prime minimax-h3 grok-imagine-video-1.5

Webhooks: get notified when a job finishes

Video and image generation are asynchronous: you submit, get a job ID, and the result arrives a little later. Instead of polling, let us push the result to your server when the job ends, and warn you when your balance runs low.

Rather not poll? Add callback_url when you submit and we push the result to your server when the job ends, or set a default endpoint in the console. Webhook docs →

Agent Skills: make a video in one sentence from your coding agent

Install the NezhaGate skill in Claude Code, Codex, Cursor or another skill-aware agent, then just ask: "turn this image into an 8-second video with seedance-2.5 and save it to ./out". The agent picks the model, uploads your files, waits for the render and saves the result. Available now: the Seedance 2.5 / 2.0 video skill.

npx skills add https://nezhagate.com

export NEZHAGATE_API_KEY=YOUR_API_KEY

github.com/gaoorange/nezhagate-skills · Demo clip (8 s, one request)

Integration Notes to Copy for AI

Send the whole block below to Codex / Claude Code / Cursor and it will know how to integrate. You can also use “Copy Page” in the top right to copy the entire doc.

Integrate NezhaGate through its OpenAI-compatible API.
Base URL: https://nezhagate.com/v1
API Key: read it from the environment variable NEZHAGATE_API_KEY.
Claude models can also use the native Anthropic protocol (for Claude Code / the Anthropic SDK): ANTHROPIC_BASE_URL=https://nezhagate.com/anthropic, headers x-api-key=API Key and anthropic-version: 2023-06-01, POST /anthropic/v1/messages (max_tokens is required), models claude-sonnet-4-6 / claude-opus-5 / claude-fable-5 / claude-opus-5-5 / claude-sonnet-5-5.
Chat models (/chat/completions): gpt-5.6-sol / gpt-5.6-luna / gpt-6-astra / gpt-6.1-sol / gpt-6-sol / gpt-6-luna / claude-sonnet-4-6 / claude-opus-5 / claude-fable-5 / claude-opus-5-5 / claude-sonnet-5-5 / gemini-4-argon / gemini-3.1-pro / gemini-3.8-flash / deepseek-v4.1-flash / deepseek-v4-flash-0731 / glm-5.3 / qwen3.8-max / qwen3.8-flash / grok-4.7
Image models (text-to-image /images/generations; image-to-image /images/edits or add an image parameter): gpt-image-2 / gpt-image-2.5-flare / gpt-image-2.5-sunburst / nano-banana-2 / nano-banana-pro / nano-banana-2.1 / grok-imagine-image / grok-imagine-image-quality
Video models (/videos/generations, async jobs; add reference images for image-to-video; how each model uses them is in its own doc): veo-3.1 / gemini-omni-flash / kling-4.0 / seedance-2.5-token / seedance-2.5-token-draft / seedance-2.5 / seedance-2.0 / seedance-2.0-fast / seedance-2.0-clip / seedance-2.5-30s / wan3.0-video / wan3.0-video-prime / minimax-h3 / grok-imagine-video-1.5
Auth: Authorization: Bearer YOUR_API_KEY
Requirements (chat): return choices[0].message.content; on 401 check the key, 402 means insufficient balance, 429/5xx can be retried.
Requirements (images, async): POST /images/generations or /images/edits returns HTTP 202 with a job id right away ({"id":"img_...","object":"image.generation.job","status":"queued"}); then poll GET /v1/images/jobs/{id} until status=succeeded and read data[0].url (an img.nezhagate.com link kept for 15 days; download it or copy it to your own storage). Do not read data right after submitting.
Requirements (video, async): POST /videos/generations returns HTTP 202 with a job id (object=video.generation.job); then poll GET /v1/videos/jobs/{id} until status=succeeded and read data[0].url (a re-hosted mp4 link kept for 15 days). Render time varies by model (from a minute or two to over an hour); failed jobs are refunded in full.

Check balance & usage

Use any API key to read the account remaining balance and spend — no dashboard login needed. Simplest: GET /v1/usage. Also OpenAI-compatible: /v1/dashboard/billing/credit_grants (balance = total_available).

curl https://nezhagate.com/v1/usage -H "Authorization: Bearer YOUR_API_KEY"

curl https://nezhagate.com/v1/dashboard/billing/credit_grants -H "Authorization: Bearer YOUR_API_KEY"

Error Codes

Errors from the /v1 endpoints share the shape below (the /anthropic endpoints use Anthropic's own format). code is a stable, machine-readable identifier: branch on code, never on the message text.

{
  "error": {
    "message": "human readable",
    "type": "invalid_request_error | authentication_error | permission_error | insufficient_quota | rate_limit_error | server_error",
    "code": "stable_snake_case",
    "param": "field_name_or_null"
  }
}

Full list of error codes →

max_tokens on thinking models — Gemini 3.x Flash (3-flash-preview / 3.6 / 3.7 / 3.8, every tier) spend the budget on hidden reasoning before writing the visible answer, so a small budget is consumed before the answer starts: measured, max_tokens=16 fails outright and 32 / 64 return a plausible-looking wrong value with finish_reason=length. The gateway raises the output budget for those models to at least 512 (you are billed only for tokens actually used), but send >=512 yourself; treat any finish_reason=length as possibly truncated and retry with a larger max_tokens.

Common questions on access, billing, models and safety. NezhaGate FAQ →