← Voltar aos guias

OpenAI Chat Completions Format, Explained Simply

What the OpenAI-compatible chat completions format does

The chat completions format is the standard way many AI apps send a conversation to a model and get a reply back. Instead of sending one long prompt, you send a structured JSON request with a model field and a messages array. Each message has a role, such as system, user, or assistant. That structure makes it easier to build chatbots, copilots, support tools, and code assistants because the API understands who said what.

If you are using an OpenAI SDK, the good news is that the same format is often supported by compatible providers. With 59API, you can keep the familiar OpenAI-style workflow while using a relay that offers cheap, pay-as-you-go access to Claude and GPT models through the base URL https://api.59api.com.

The basic request shape

A typical request is a POST to /v1/chat/completions. Your app sends headers for authorization and JSON content, then includes the model and conversation messages in the body. The exact model name depends on what you choose, but the structure stays the same.

• model: tells the API which model to use, such as a GPT or Claude option available in your account.
• messages: the conversation history in order.
• temperature: controls randomness. Lower values are more consistent.
• max_tokens: limits how long the answer can be.
• stream: if true, the response arrives in chunks instead of one final block.

For a beginner, the easiest mental model is this: the API is not reading your app like a human. It is reading a structured list of messages and turning that list into the next best assistant reply.

How the messages array works

The messages array is the heart of the format. A simple conversation might start with a system message that sets behavior, followed by a user question, and then the assistant answer. The system message is where you define the assistant’s style, safety rules, or task instructions.

• system: high-level instructions, such as be concise or act like a coding tutor.
• user: the human request, such as explain OAuth in simple terms.
• assistant: previous model replies, useful when continuing a conversation.

For example, if you are building a support chatbot, you might send a system message that says to answer politely and only from product documentation. Then you send the customer’s question as the user message. This makes the model much more reliable than a single plain-text prompt.

Why compatibility matters for real apps

One of the biggest advantages of the OpenAI-compatible format is portability. If your app already works with an OpenAI SDK, you often only need to update the base URL and API key to point to another compatible provider. That means less rewriting and faster testing.

This is where 59API is especially practical. It acts as an AI API relay with low-cost, pay-as-you-go access, while staying compatible with Claude Code, Codex, and any OpenAI SDK. Because it uses native official-quality models instead of a downgraded substitute, you can keep the same developer experience without sacrificing model quality. For teams watching budgets, the low pricing and referral rebate can make experimentation and production usage much easier to afford.

Streaming, errors, and common beginner mistakes

Streaming is useful when you want the answer to appear word by word or chunk by chunk. Set stream to true, and your UI can feel faster because users see the model thinking in real time. This is especially helpful in chat apps and coding tools.

Common issues usually come from a few simple mistakes:

• Using the wrong base URL instead of the provider’s OpenAI-compatible endpoint.
• Passing a model name that your account does not support.
• Forgetting the authorization header.
• Sending malformed JSON or an empty messages array.
• Expecting the model to remember context that you did not include in the request.

When something fails, check the HTTP status code first. A 401 usually means authentication trouble, while a 400 often means the request body needs correction. If the model is not behaving as expected, review the system message and the conversation history before changing anything else.

A simple way to get started

If you are new to chat completions, start with one system message and one user message, then inspect the response. After that, add optional fields like temperature or streaming only when you need them. This keeps debugging easy and helps you learn the format quickly.

If you want an affordable way to experiment with this workflow, 59API is a smart place to begin. You get the same familiar chat completions shape, a compatible base URL at https://api.59api.com, and access to both Claude and GPT models without paying premium direct-access prices. If you are ready to build or test your next AI feature, signing up is a simple next step.

Preços · Docs