← Voltar aos guias

Call Claude and GPT from Python with One API

Why use one Python client for Claude and GPT?

Most AI applications eventually need more than one model. You may use a fast, lower-cost model for classification, a stronger model for code review, and a premium model for difficult reasoning. Maintaining separate providers, SDKs, billing accounts, and request formats adds unnecessary operational work.

59API provides pay-as-you-go access to native official-quality Claude and GPT models through an OpenAI-compatible API. That means one Python integration can route requests to Claude Opus, Sonnet, Haiku, Fable, and GPT models by changing the model name. It is also compatible with tools built around the OpenAI SDK, including Codex and Claude Code workflows. For teams watching spend, 59API is a practical low-cost relay option without deliberately downgrading model quality.

1. Create an API key and install the SDK

Sign up for a 59API account, add a small pay-as-you-go balance, and create an API key in the dashboard. Keep the key out of source control. The service also offers a referral rebate, which can reduce effective costs for developers who share it with their network.

Install the official OpenAI Python package in a virtual environment:

python -m venv .venv

source .venv/bin/activate

pip install openai

Set the secret in your shell or deployment platform. On macOS or Linux, use export API_KEY="your_59api_key". In production, store it in a secret manager or the platform's encrypted environment-variable settings.

2. Configure Python for 59API

Create a file named app.py. The OpenAI SDK accepts a custom base URL, allowing it to send the same chat-completions request format through 59API. Use the API endpoint path shown in the 59API dashboard; for OpenAI-compatible clients, this is commonly the base URL plus /v1.

from openai import OpenAI

import os

client = OpenAI(
api_key=os.environ["API_KEY"],
base_url="https://api.59api.com/v1",
)

Before hard-coding a model identifier, check the current 59API model list in the dashboard or documentation. Providers can update aliases and availability. This is especially important when selecting among Claude Opus, Sonnet, Haiku, Fable, and GPT variants.

3. Send your first model request

Use a system message to define the application's rules and a user message for the task. The following example uses a placeholder model alias. Replace it with an available Claude or GPT model name from your 59API account.

response = client.chat.completions.create(
model="your-model-name",
messages=[
{"role": "system", "content": "You are a concise Python assistant."},
{"role": "user", "content": "Explain a Python context manager in two sentences."},
],
temperature=0.2,
)

print(response.choices[0].message.content)

To call another model, retain the same client and messages, then change only model. For example, route inexpensive bulk summarization to a fast model and reserve a larger Claude or GPT model for complex analysis. Keeping prompts and output handling consistent makes A/B testing far easier.

4. Make model selection deliberate

• Fast or low-cost models: Use them for extraction, tagging, short summaries, and routine customer-support drafts.
• Stronger models: Use them for multi-step reasoning, code generation, long-document analysis, and high-impact answers.
• Deterministic tasks: Set a low temperature, usually between 0 and 0.3, for structured outputs and repeatable behavior.
• Creative tasks: Increase temperature carefully, then evaluate outputs against examples before release.

Do not select a model solely by its advertised capability. Measure latency, error rate, output quality, and token usage against representative production prompts. A smaller model with a precise prompt is often the best economic choice.

5. Add production safeguards

Wrap API calls with retries for transient network failures and rate-limit responses. Log request IDs, model names, elapsed time, and token usage, but never log API keys or sensitive user content. Set application-level timeouts, cap input length, and validate any structured response before passing it to downstream systems.

For workloads where prompts are similar, cache results using a hash of the normalized prompt, model, and relevant settings. For user-facing applications, show a useful fallback message when an upstream request fails. Keep a second approved model available so critical features can continue when a specific model is unavailable.

59API lets you start with this single OpenAI SDK integration while choosing Claude or GPT according to the task and budget. Create an account, verify the current model aliases, and run a small benchmark with your real prompts before moving traffic into production.

Preços · Docs