Pay-as-you-go API
OpenAI- and Anthropic-compatible. Prepaid balance, billed on actual tokens. Works with any client or program that supports a custom endpoint.
Three steps
Add funds in the console. Credited 1:1, never expires.
Generate an sk-neox- key, optionally scoped to models and a spend limit.
Point your client or SDK at the base URL below.
https://gateway.neox-dev.com/v1from openai import OpenAI
client = OpenAI(
base_url="https://gateway.neox-dev.com/v1",
api_key="sk-neox-...",
)
reply = client.chat.completions.create(
model="deepseek-v4-flash",
messages=[{"role": "user", "content": "hi"}],
)
print(reply.choices[0].message.content)Models & pricing
Billed on actual tokens, USD per million tokens. Failed requests are not charged.
Loading…
Errors and limits
Errors come back in an OpenAI-compatible body: {"error": {"code", "message", "requestId"}}. Branch on error.code, not on the message text.
platform.invalid_keyInvalid or revoked keyCheck the key or create a new one in the consoleplatform.insufficient_balanceBalance too lowTop up in the console and retryplatform.key_limitThis key hit its spend limitRaise the key limit or use another keyplatform.wallet_frozenBalance is frozenContact support@neox-dev.complatform.model_not_offeredModel not offeredList models with GET /modelsquota.rate_limitRate limited: 600 requests/min and 10 concurrent per accountLower concurrency and retry with backoffupstream.rate_limitedThe model provider is throttlingRetry later or switch modelsHow it differs from Neox Studio
A subscription for the Neox desktop app, CLI and Android, with an included allowance.
Pay as you go, from any client or your own programs. Separate from your subscription.
FAQ
Does my balance expire?
No. Top-ups are credited 1:1 and drawn down as you use them.
Am I charged for failed requests?
No. A hold estimated from the input and the maximum output (max_tokens) is placed when a request starts, settled on actual tokens when it succeeds, and released in full if it fails.
Which endpoints are supported?
OpenAI-compatible /v1/chat/completions (including streaming), /v1/responses, /v1/embeddings and /v1/models, plus Anthropic-compatible /v1/messages (including streaming and tool use).
Can I cap what a key spends?
Yes. Each key can have its own spend limit and allowed models; it is refused once the limit is reached.
Do requests pass through Neox servers?
Yes. Platform requests are relayed to model providers through the Neox gateway. Prompt content is not retained by default.
Is the Anthropic format (/v1/messages) supported?
Yes. Set the base URL of your Anthropic SDK or tool to https://gateway.neox-dev.com (without /v1) and use your sk-neox- key. Both x-api-key and Authorization: Bearer are accepted. /v1/messages/count_tokens returns an estimate.
Can I run my own agent on it?
Yes. The open-source Neox Agent SDK connects with an openai-compatible provider. See the Neox Agent SDK tab under Examples, and /sdk.
Can my platform balance be used in the Neox apps?
Yes. Your account has a single platform balance: besides API calls, subscribers who use up their plan allowance in the Neox desktop app, CLI or Android can keep going on the balance (extra usage, which you can turn off anytime).