AICU LLM API Documentation

OpenAI-compatible chat completions API. An API key (Bearer auth) is required. Issue an aicu_live_ key from the dashboard.

Base URL

https://api.aicu.ai/v1

Quick Start

curl -X POST https://api.aicu.ai/v1/chat/completions \
  -H "Authorization: Bearer $AICU_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v3",
    "messages": [{"role": "user", "content": "Hello!"}],
    "max_tokens": 100
  }'

🖼️ Image input (Vision) works too. Make content an array and pass an image_url → Vision guide

POST /v1/chat/completions

OpenAI-compatible chat completions endpoint.

Request:

{
  "model": "deepseek-v3",
  "messages": [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "Hello!"}
  ],
  "max_tokens": 1000,
  "temperature": 0.7,
  "stream": false
}

Parameters:

Name	Type	Required	Description
model	string	Yes	Model ID
messages	array	Yes	Array of messages
max_tokens	number	No	Maximum tokens to generate (default: 4096)
temperature	number	No	Sampling temperature, 0-2 (default: 1)
stream	boolean	No	Stream the response (default: false)

Response:

{
  "id": "gen-xxx",
  "object": "chat.completion",
  "created": 1234567890,
  "model": "deepseek-v3",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Hello! How can I help you today?"
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 10,
    "completion_tokens": 15,
    "total_tokens": 25
  }
}

Response Headers:

X-AICU-Model: Model used
X-AICU-Provider: Provider (openrouter/groq/aicu)
X-AICU-Tokens: Tokens consumed
X-AICU-AP-Cost: AP charged
X-AICU-Latency-Ms: Latency

GET /v1/models — every model, one list

GET /v1/models is the single source of truth for everything the platform serves: chat, image, speech (tts) and transcription (stt). Every entry has the same shape — type, the endpoint to call, stage, pricing with an explicit unit, docs_url, pricing_url and updated_at — and the list envelope carries docs, pricing_url and the latest updated_at. GET /v1/models/:id returns one model (OpenAI-style retrieve; image aliases such as flare work).

curl https://api.aicu.ai/v1/models            # everything
curl https://api.aicu.ai/v1/models/deepseek-v3

{
  "object": "list",
  "docs": "https://api.aicu.ai/docs",
  "pricing_url": "https://api.aicu.ai/docs/pricing",
  "updated_at": "2026-09-11",
  "data": [
    { "id": "deepseek-v3", "type": "chat", "endpoint": "POST /v1/chat/completions", "stage": "beta",
      "pricing": { "unit": "per_1k_tokens", "ap_in_per_1k": 9, "usd_in_per_1k": 0.0009, "ap_out_per_1k": 27, "usd_out_per_1k": 0.0027 },
      "docs_url": "https://api.aicu.ai/docs/llm-api", "updated_at": "2026-08-01" },
    { "id": "gpt-image-2.5-flare", "type": "image", "endpoint": "POST /v1/images/generations",
      "pricing": { "unit": "per_image", "ap_from": 700, "ap_by_size_quality": { "1024x1024": { "low": 700, "medium": 1600, "high": 6400 } } } }
  ]
}

stage is beta until production starts on 2026-09-23, then production; preview marks models we may still change.

GET /v1/chat/models

The same array filtered to type=chat. Kept for compatibility; the fields below are unchanged.

curl https://api.aicu.ai/v1/chat/models

Response:

{
  "object": "list",
  "data": [
    {
      "id": "deepseek-v3",
      "owned_by": "openrouter",
      "pricing": {"ap_per_1k_tokens": 14, "ap_in_per_1k": 9, "ap_out_per_1k": 27},
      "stage": "beta",
      "description": "汎用、コスパ良好"
    }
  ]
}

Available Models

Pricing tiers:

🟢 Free — permanently free (rate limits apply)
🟡 Free for a limited time — α / preview stage. No real billing through 2026-10-31 (Kumamoto Earthquake 2026 relief program). The AP figures in the table are the rates planned for after GA
💠 Paid (planned rates) — the AP figures are reference rates that apply from GA onward

The platform as a whole is currently in alpha. None of the tiers above bill end users today (usage is still logged). Rates are given as "input / output AP per 1K tokens". 1 USD = 10,000 AP (1 AP = $0.0001).

:::caution Amounts are approximate and rates get revised The AP figures below and every amount in this document are approximate (AP→USD is fixed at 10,000 AP = $1, but yen conversion moves with the exchange rate). Per-model AP rates are subject to price revisions. This table is maintained by hand, so always base estimates and invoices on the current values from GET /v1/chat/models. :::

🟢 Free (Groq — stays free after GA)

Model	Provider	Rate	Best for
`groq-llama-3.3-70b`	Groq	Free	High quality, rate-limited
`groq-llama-3.1-8b`	Groq	Free	Fast, rate-limited

🟡 Free for a limited time (α / preview — through 2026-10-31)

Model	Provider	in / out (AP/1K)	Best for
`gpt-4o`	OpenAI	76 / 300	High-quality multimodal
`gpt-4o-mini`	OpenAI	5 / 18	Lightweight and fast
`gpt-5.6-sol`	OpenAI	150 / 900	Top of the line. Hard reasoning, long context
`gpt-5.6-terra`	OpenAI	80 / 450	Half the price of sol. The everyday workhorse
`gpt-5.6-luna`	OpenAI	30 / 180	Light and fast. Chat and summarization
`gpt-5.4-mini`	OpenAI	30 / 140	Inexpensive. Routine generation and extraction
`gpt-5.4-nano`	OpenAI	6 / 40	Cheapest. Short jobs like classification and scoring
`sakura-kimi`	Sakura AI Engine	12 / 60	Kimi-K2.6, inferred in Japan. Your data is not used for training
`kimi-k2.7-code`	Sakura AI Engine	11 / 101	Coding-focused. OpenCode / Cline compatible

💠 Paid (beta — planned rates)

The Gemini models moved to the rates below on 2026-08-01: the default gemini-flash now maps to the lighter 2.5 Flash-Lite, and a new gemini-3.6-flash covers the latest model.

Model	Provider	in / out (AP/1K)	Best for
`deepseek-v3`	OpenRouter	9 / 27	General-purpose, good value
`gemini-flash`	OpenRouter	3 / 12	Lightweight and cheap (Gemini 2.5 Flash-Lite). From 2026-08-01
`gemini-3.6-flash`	OpenRouter	45 / 225	Latest and highest quality (Gemini 3.6 Flash). Added 2026-08-01
`llama-3.1-8b`	OpenRouter	1 / 2	Lightweight
`llama-3.1-70b`	OpenRouter	4 / 10	High quality
`qwen3-32b`	OpenRouter	4 / 10	Large context

The previous generation — gpt-5 / gpt-5-mini / gpt-5-nano — stays available for now, but new work should target the gpt-5.6 / gpt-5.4 families. The always-current list is GET /v1/chat/models; this table is updated by hand and may lag behind.

Notes on reasoning models

Models that GET /v1/chat/models reports with reasoning: true (the gpt-5 family and similar) bill thinking tokens as well, and a small max_tokens can be consumed entirely by reasoning, leaving an empty body (content: "").

Pass reasoning_effort (minimal / low / medium / high) to control how much the model thinks — it is forwarded as-is, OpenAI-compatible.
When it is omitted for a gpt-5-family model, api.aicu.ai applies a default of minimal so that even short prompts come back with a body. An explicit value always wins.
For short generations where cost and predictability matter more, a non-reasoning model such as gpt-4o-mini is the reliable choice.
What the levels mean upstream: OpenAI — Reasoning. With tools in the request the effort is dropped, because reasoning and tool calling cannot both run — see Coding agents.

curl -X POST https://api.aicu.ai/v1/chat/completions \
  -H "Authorization: Bearer $AICU_API_KEY" -H "Content-Type: application/json" \
  -d '{"model":"gpt-5-mini","messages":[{"role":"user","content":"140字で告知文を書いて"}],"max_tokens":600,"reasoning_effort":"minimal"}'

OpenAI SDK Compatibility

from openai import OpenAI

client = OpenAI(
    base_url="https://api.aicu.ai/v1",
    api_key="aicu_live_xxx"  # API key required (issue one in the dashboard)
)

response = client.chat.completions.create(
    model="deepseek-v3",
    messages=[{"role": "user", "content": "Hello!"}]
)
print(response.choices[0].message.content)

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://api.aicu.ai/v1',
  apiKey: 'aicu_live_xxx', // API key required (issue one in the dashboard)
});

const response = await client.chat.completions.create({
  model: 'deepseek-v3',
  messages: [{ role: 'user', content: 'Hello!' }],
});

Where the upstream specification lives

api.aicu.ai is OpenAI-compatible, which is only a useful promise if you can see what it is compatible with. These are the upstream documents this API follows. When our page and the upstream page disagree about a parameter, the upstream page describes the parameter and this site describes what we accept — if you hit a difference, tell us at aicu.ai/contact.

Topic	Upstream reference	How we differ
Request and response shape	OpenAI — Create chat completion	Same shape. Extra `X-AICU-*` response headers carry AP cost, model and latency
Tool / function calling	OpenAI — Function calling	Forwarded as-is. With a reasoning model we drop the reasoning effort so the tool call succeeds (see Coding agents)
Reasoning models, `reasoning_effort`	OpenAI — Reasoning	Forwarded as-is. We default to `minimal` when you omit it, so short prompts still return a body
Image input	OpenAI — Images and vision	Same shape. Which models accept images is measured on our side — see Vision
Groq-hosted models	Groq — OpenAI compatibility	Groq's own limits apply on top of ours

What is ours, not upstream: AP billing and the X-AICU-AP-Cost header, the model catalogue (GET /v1/chat/models), key scopes, and the hold-then-settle charging described in Pricing. Those are not in any upstream document.

Credits (billing)

10,000 AP = $1 (USD-denominated; 1 AP = $0.0001. Our own costs are in USD, so the unit follows)
Yen figures are approximate and move with the exchange rate — treat every amount in these docs as an estimate
Per-model AP rates are subject to price revisions. See GET /v1/chat/models for current values
Groq models: free (rate limits apply)
Every endpoint requires an API key (Bearer auth)

Rate Limits

Provider	Limit
Groq	30 RPM
OpenRouter	Depends on your plan

Troubleshooting

A 403 is about your key, not your HTTP client

Until 2026-08-23, clients that kept their default User-Agent — Python's standard-library urllib was the usual suspect — could be blocked as bots and get a 403 (Cloudflare error 1010). That is fixed. api.aicu.ai is exempt from the bot check, and a default urllib request now returns 200 (verified 2026-09-22). You do not need to set a User-Agent.

If you still get a 403, it is about the key itself:

Body	Meaning
`{"error": "API key does not have required scope: llm", "required_scope": "llm", "current_scopes": [...]}`	The key lacks that surface. Add it with 権限変更 at dashboard/keys — no reissue needed
`{"error": {"code": "unauthorized", ...}}`	The key is revoked or does not exist

Support

Skill: https://api.aicu.ai/skills/llm
Dashboard: https://api.aicu.ai/dashboard
Contact: https://aicu.ai/contact

Base URL​

Quick Start​

POST /v1/chat/completions​

GET /v1/models — every model, one list​

GET /v1/chat/models​

Available Models​

🟢 Free (Groq — stays free after GA)​

🟡 Free for a limited time (α / preview — through 2026-10-31)​

💠 Paid (beta — planned rates)​

Notes on reasoning models​

OpenAI SDK Compatibility​

Where the upstream specification lives​

Credits (billing)​

Rate Limits​

Troubleshooting​

A 403 is about your key, not your HTTP client​

Support​