GPT on RunAPI

Use the official OpenAI SDK (Python, TypeScript, Ruby) — or any OpenAI-compatible HTTP client — and switch the base URL to https://runapi.ai/v1. The endpoints speak the standard OpenAI protocol: Chat Completions (POST /v1/chat/completions) and the Responses API (POST /v1/responses). No client code changes beyond base_url and api_key.

Setup

OPENAI_API_KEY=YOUR_RUNAPI_TOKEN
OPENAI_BASE_URL=https://runapi.ai/v1

Get a RunAPI API Key at \x3Chttps://runapi.ai/api_keys>.

Pick the right endpoint

Core recipe — Chat Completions

from openai import OpenAI

client = OpenAI(api_key="YOUR_RUNAPI_TOKEN", base_url="https://runapi.ai/v1")

response = client.chat.completions.create(
    model="gpt-5.4",
    messages=[{"role": "user", "content": "Explain quantum computing simply."}],
    reasoning_effort="high",
)
print(response.choices[0].message.content)
print(response.usage)

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: "YOUR_RUNAPI_TOKEN",
  baseURL: "https://runapi.ai/v1",
});

const response = await client.chat.completions.create({
  model: "gpt-5.4",
  messages: [{ role: "user", content: "Explain quantum computing simply." }],
});

Core recipe — Responses API

import httpx

response = httpx.post(
    "https://runapi.ai/v1/responses",
    headers={"x-api-key": "YOUR_RUNAPI_TOKEN"},
    json={
        "model": "gpt-5.4",
        "input": "Explain the theory of relativity.",
        "reasoning": {"effort": "medium"},
    },
)
print(response.json())

The Responses API takes input (string or structured), reasoning.effort ("low" / "medium" / "high"), and optional include for thinking blocks.

Streaming

stream = client.chat.completions.create(
    model="gpt-5.4",
    messages=[{"role": "user", "content": "Write a haiku about coding."}],
    stream=True,
)
for chunk in stream:
    delta = chunk.choices[0].delta.content
    if delta:
        print(delta, end="", flush=True)

const stream = await client.chat.completions.create({
  model: "gpt-5.4",
  messages: [{ role: "user", content: "Write a haiku about coding." }],
  stream: true,
});

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0].delta.content ?? "");
}

Streaming runs through a regional edge proxy so the request does not hold a Rails/Puma thread. Long generations should always stream.

Vision / multimodal

{
  "model": "gpt-5.4",
  "messages": [
    {
      "role": "user",
      "content": [
        { "type": "text", "text": "What is in this image?" },
        { "type": "image_url", "image_url": { "url": "https://example.com/img.jpg" } }
      ]
    }
  ]
}

Standard OpenAI multimodal block — works on both Chat Completions and Responses (Responses also accepts structured input items).

Tool use / function calling / web search

{
  "model": "gpt-5.4",
  "messages": [
    { "role": "user", "content": "Find the latest news on RunAPI." }
  ],
  "tools": [
    { "type": "function", "function": { "name": "web_search" } }
  ]
}

web_search is supported across the GPT models above. Custom function tools use the standard OpenAI tools schema.

List models

curl https://runapi.ai/v1/models -H "Authorization: Bearer YOUR_RUNAPI_TOKEN"

Returns OpenAI-compatible model objects. If the API Key has allowed_models restrictions, only permitted models are returned.

Supported models

Connect Codex CLI itself

export OPENAI_BASE_URL=https://runapi.ai/v1
export OPENAI_API_KEY=YOUR_RUNAPI_TOKEN
codex

Agent rules

• Pro models (gpt-5.*-pro) reject Chat Completions — always use Responses for them. Other models accept either endpoint.
• Use streaming for any response longer than a few hundred tokens. Do not hold the agent on a long blocking request.
• reasoning_effort is supported on every GPT model above; default is usually "high" for non-Pro models.
• Pricing, rate limits, quotas — link to \x3Chttps://runapi.ai/models/gpt.md>, not this skill file.

Routing

• Model page: \x3Chttps://runapi.ai/models/gpt.md>
• Provider page: \x3Chttps://runapi.ai/providers/openai.md>
• Catalog: \x3Chttps://runapi.ai/models.md>