How to use your API subs key

Three things: the base URL, your API key, and a model from your key's tier.

Base URL · OpenAI, Codex, Cursor, n8n…	https://api.aisubs.id/v1
Base URL · Claude Code (no /v1)	https://api.aisubs.id
API key	sk-… from your seller or your API subs account

Models per tier

One key = one tier; inside it you can switch models freely. A GPT key can't call Claude models and vice versa.

Claude Max route
claude-opus-5-5 · claude-sonnet-5 · claude-opus-5 · claude-haiku-4-5 · claude-opus-4-8 · claude-sonnet-4-6
ChatGPT Plus route
gpt-6-sol · gpt-6.1-sol · gpt-6-astra · gpt-5.6-sol · gpt-5.6-terra
Official OpenAI relay
gpt-6-sol · gpt-6.1-sol · gpt-6-astra · gpt-5.6-sol · gpt-5.6-terra
ChatGPT Pro route
gpt-6-sol · gpt-6.1-sol · gpt-6-astra · gpt-5.6-sol · gpt-5.6-terra
Claude Kiro
claude-opus-5-5 · claude-sonnet-5 · claude-opus-5 · claude-haiku-4-5 · claude-opus-4-8 · claude-sonnet-4-6
SuperGrok Heavy
grok-4.7 · grok-4.6 · grok-4.5
Gemini Antigravity
gemini-3.8-flash · gemini-2.5-flash
Chinese Models
deepseek-v4.1-flash · deepseek-v4-pro · kimi-k3-1 · kimi-k2.8-preview · kimi-k2.7 · kimi-k2.6 · glm-5.3 · glm-5.2 · minimax-m3

Images

Billed per finished image. Hanging up halfway still costs one image (it is still made upstream).

GPT Image
gpt-image-2 · gpt-image-2.5 · gpt-image-2.5-flare · gpt-image-2.5-sunburst
Endpoint: POST /v1/images/generations (also /v1/images/edits)
Nano Banana
nano-banana-2 · nano-banana-2.1 · nano-banana-pro
Endpoint: POST /v1beta/models/{model}:generateContent (Gemini format, key in x-goog-api-key)

curl https://api.aisubs.id/v1/images/generations \
  -H "Authorization: Bearer $AISUBS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-image-2", "prompt": "a blue robot mascot", "size": "1024x1024"}'

How the multiplier works

Every tier has a multiplier. Your cost = the model's official price × the multiplier.
Example: the Claude Max route has multiplier 0.26 — Claude Opus 5.5 officially costs $4 per 1M input tokens, so you pay $4 × 0.26 = $1.04.
The ChatGPT Plus route is 0.0089 — gpt-6-sol input costs $2 × 0.0089 = $0.0178 per 1M tokens.
$1 of API credit = $1. You only pay for tokens actually used: input, cache and output each at their own price.
Official / Pure labels: Official = OpenAI's official relay; Pure = the ori/pure model. A yellow "Disrupted" label means that model is down right now.

Use the base URL and your key in any of these tools.

Codex CLI

Add a provider to ~/.codex/config.toml, then set the key in your shell:

# ~/.codex/config.toml
model = "gpt-6-sol"
model_provider = "aisubs"

[model_providers.aisubs]
name = "AISUBS"
base_url = "https://api.aisubs.id/v1"
env_key = "AISUBS_API_KEY"
wire_api = "responses"

Claude Code

Use a key from the Claude tier. Set these before starting claude:

export ANTHROPIC_BASE_URL="https://api.aisubs.id"
export ANTHROPIC_AUTH_TOKEN="sk-..."
export ANTHROPIC_MODEL="claude-sonnet-5"
claude

Cursor / Cline / Continue / OpenCode

Choose "OpenAI compatible" as the provider, then fill in:

Base URL : https://api.aisubs.id/v1
API Key  : sk-...
Model: any name from the Models tab, e.g. gpt-6-sol

n8n

Add an OpenAI node → Credentials → new OpenAI credential:

API Key  : sk-...
Base URL : https://api.aisubs.id/v1
Model: any name from the Models tab, e.g. gpt-6-sol

OpenAI SDK (Python / JS)

curl https://api.aisubs.id/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-..." \
  -d '{
    "model": "gpt-6-sol",
    "messages": [{"role": "user", "content": "Halo! Perkenalkan dirimu."}]
  }'

Quick test

Paste in a terminal (replace sk-…). A reply means your key works.

curl https://api.aisubs.id/v1/chat/completions \
  -H "Authorization: Bearer sk-..." \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-6-sol","messages":[{"role":"user","content":"Halo"}]}'

Good to know

Usage is charged per request, by the tokens actually used. When a key's limit runs out it stops; ask your seller to top up.
Coding agents are much cheaper than plain chat because repeated context is read from cache.
GPT requests over 272K tokens count input and cache at 2×.
For long tasks use streaming so the connection doesn't drop.
404 "model not supported" = wrong model name or not in your key's tier.