Baidu Qianfan

What it does

Baidu Qianfan takes in text and multimodal requests sent through its API, along with model parameters such as temperature and top_p, and returns model-generated responses. Developers can also use it to build Agent-based applications that combine models with tools and MCP connections. The platform accepts chat messages in an OpenAI-compatible format and returns completions, with options for structured output, function calling, and visual understanding.

People mainly use Qianfan by creating an API key in the Baidu AI Cloud console and sending requests over HTTP or through the official Qianfan SDK or an OpenAI-compatible SDK. The product page describes an all-in-one approach that combines model access, Agent development, and enterprise features, with built-in debugging and logging. Its distinguishing point is access to ERNIE models alongside several open-source model families from one endpoint.

Qianfan suits developers and enterprise teams building text, chat, or multimodal applications who want one API for several model families. Access requires a Baidu AI Cloud console account and an API key. Per-token model prices are listed on a campaign page, but regional availability and account eligibility are not confirmed. Developers should verify model availability and pricing in the official console before committing to production use.

What it can do

• Text generation and conversational model access through a unified API
• Multimodal input including visual understanding for image and video analysis
• Structured output and function calling for application integration
• Access to ERNIE and open-source foundation models such as DeepSeek, Minimax, Kimi, and GLM
• OpenAI-compatible RESTful API and multi-language SDKs with built-in debugging and logging
• Agent development support combining model services with tools and MCP

Works with

• OpenAI-compatible SDK

Example workflows

Chat completion from an application

Illustrative example. Input: a developer's user message and a system prompt. Steps: create an API key, select a model name from the model list, send a chat completions request to the Qianfan endpoint with the bearer key. Output: the model's text response returned in the completion object.

Agent application with tool connections

Illustrative example. Input: a task description and a set of tools exposed through MCP or function definitions. Steps: configure the Agent with a model, register the tools, and send requests that allow the model to call them. Output: model responses that incorporate tool results, as described in the product page's Agent engine and tools section.

Before you choose

• Model prices shown on the campaign page are per 1 million tokens, with separate input, cached input, and output rates. Actual pricing may vary by model version, region, and billing method.
• Prices and model availability should be confirmed for your region and account before production use.
• The product page describes enterprise-grade security and data isolation; confirm the specific compliance terms in the official documentation before handling regulated data.

Listed under

• Coding
• Sandboxes and infrastructure
• AI platform
• Cloud
• Consulting
• Enterprise
• Infrastructure

Model API pricing is charged per token with separate input, cached input, and output rates; the rate table on the campaign page shows examples only, and no free tier or trial period was confirmed.

Source: intl.cloud.baidu.com/en/product/qianfan_campaign.html, checked 29 Oct 2026. Prices can change; check before you buy.