OpenAI SDK for Language Translation

Translation pipelines are moving from proprietary APIs to open-source models, but most inference providers still bill by the token. For language translation, that is a structural problem. A single document can contain thousands of tokens, and multilingual prompts often carry extra overhead from Unicode and formatting. When cost scales linearly with input length, long-form translation becomes expensive to run at scale. Oxlo.ai uses a request-based pricing model: one flat cost per API call regardless of prompt length. That makes it a natural fit for translation workloads where context windows are large and predictability matters.

Why Token-Based Billing Breaks Translation Workflows

Most inference platforms meter input and output tokens separately. A 5,000-word legal contract or technical manual can balloon to 8,000 to 10,000 tokens before the model even begins translating. If you are processing batches of documents or maintaining multi-turn context for terminology consistency, token counts compound quickly. Oxlo.ai flattens that curve. Each request costs the same whether you send 100 tokens or 100,000 tokens, so translation throughput becomes a function of latency and queue depth, not a guessing game of token arithmetic.

Drop-In SDK Setup for Oxlo.ai

Oxlo.ai exposes a fully OpenAI-compatible endpoint. You only need to swap the base URL and API key. The Python SDK requires no custom adapters or wrapper classes.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.oxlo.ai/v1",
    api_key="YOUR_OXLO_API_KEY"
)

That is the entire migration. Every pattern in this guide works with your existing OpenAI SDK installation.

Basic Translation with the OpenAI SDK

A minimal translation call uses the chat completions endpoint. The key is a system prompt that locks the model into the target language and style.

response = client.chat.completions.create(
    model="qwen3-32b",
    messages=[
        {"role": "system", "content": "You are a professional translator. Translate the user's text from Spanish to English. Preserve formatting and technical terms."},
        {"role": "user", "content": "El sistema de inferencia utiliza una arquitectura de mezcla de expertos para reducir la latencia."}
    ]
)

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

Because Oxlo.ai charges per request, you can expand the system prompt with glossaries, style guides, or few-shot examples without worrying about token inflation.

Handling Long-Form and Batch Translation

For long documents, you can either chunk text or exploit large context windows. Models on Oxlo.ai such as DeepSeek V4 Flash support 1M tokens of context, letting you feed entire reports in a single request. Since the price is flat per request, a one-shot translation of a full document costs the same as a single sentence.

document = open("contract.md").read()

response = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[
        {"role": "system", "content": "Translate the following legal document from German to British English. Maintain clause numbering and defined terms."},
        {"role": "user", "content": document}
    ]
)

Batching multiple short texts into one request is another way to maximize value. Send an array of paragraphs with explicit separators and parse the structured response, or use JSON mode to enforce a machine-readable output format.

python
response = client.chat.completions.create(
    model="llama-3.3-70b",
    response_format={"type": "json_object"},
    messages=[
        {"role": "system", "content": "Translate each paragraph into French. Return a JSON object where keys are paragraph IDs and values are translations."},
        {"role": "user", "content": '{"p1": "Hello world", "p2