Online Variant

Not yet available.

The :online variant augments a model's response with real-time web search results. Before generating an answer, the model issues a search query, retrieves fresh pages, and incorporates the retrieved content. You get a grounded, up-to-date answer with source citations — no RAG pipeline or browser tool required on your end.

Usage#

curl https://opendunes.com/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENDUNES_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/gemini-2.5-pro:online",
    "messages": [{ "role": "user", "content": "What is the latest exchange rate between DZD and EUR?" }]
  }'

How the model uses the web#

OpenDunes injects a web search tool into the model's context. The model decides when to search — not every request triggers a search. If the model determines the question can be answered from its training data, it may skip the search step. Queries that are time-sensitive, location-specific, or explicitly about recent events reliably trigger a search.

The search step adds latency (typically 1–4 seconds) before the model starts generating. Use streaming ("stream": true) so the user sees the response as soon as generation begins.

Citations in the response#

When search results are used, the response content includes inline citations with numbered references:

The DZD/EUR rate today is approximately 145 DA per euro [1]. This rate has been relatively stable over the past month [2].

Sources:
[1] https://example-exchange.com/rates/dzd-eur
[2] https://example-news.com/finance/algeria-dinar

The citation format follows the model's natural output — it is part of the generated text, not a separate structured field.

Supported models#

The :online variant is available on models that support tool use (the web search capability is implemented as a tool call internally). Check the catalog for models with capability: online or filter:

GET /api/models?capability=online

Pricing (DA)#

Online requests are billed at two rates:

1. Search cost — a flat DA charge per search query issued during the request (shown in the catalog).
2. Token cost — the model's normal per-token DA rate for the full prompt (including retrieved content) and generated output.

Retrieved web content added to the context counts as prompt tokens. Long web pages in context can meaningfully increase the token count — monitor the usage field in the response.

Using with the OpenAI SDK#

import os
from openai import OpenAI

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

resp = client.chat.completions.create(
    model="google/gemini-2.5-pro:online",
    messages=[{
        "role": "user",
        "content": "What are the latest AI models released this week?",
    }],
    stream=True,
)
for chunk in resp:
    delta = chunk.choices[0].delta.content or ""
    print(delta, end="", flush=True)

Note. The :online variant is not a substitute for a full RAG pipeline over your own data. It searches the public web. For retrieval over your own documents, build a retrieval step and pass results in messages directly.