AgentKit API
火山引擎Call the Volcano Engine AgentKit agent API and get a conversational agent with tool calling and memory running in 30 minutes.
A one-stop platform for online LLM inference, fine-tuning, distillation, evaluation, and Agent development, supporting API calls and CLI access.
Baidu Qianfan is Baidu Intelligent Cloud's LLM service and Agent development platform, covering the full model lifecycle from data management, model fine-tuning, distillation, and evaluation to online invocation. Users can obtain a API Key from the console, select a model, and copy sample code to make their first call; it also supports invoking full-modal capabilities via the Qianfan CLI in terminals and local AI Agents. The platform uses pay-as-you-go postpaid billing, with separate pricing for input and output; cache hits can reduce costs, and some models are cheaper during off-peak nighttime hours. It also offers Token Plan subscription packages with a unified quota shared across multiple models.
On the model services front, the page lists multiple accessible models: ERNIE-5.5-Flash-Preview (text generation, web search), ERNIE-Light-DT-1.0 (lightweight context, max output 128K, thinking mode can be toggled on or off), ERNIE-Wizshot-ME (video generation, 1080p, 0.7 yuan/second), DeepSeek-V4.1-Flash-1017 (text generation, visual understanding, ultra-fast inference), GLM-5.3 (text generation, coding, cybersecurity), and Unlimited-OCR-2 (vision-language, long-document parsing). The end-to-end service system includes: model services (pay-as-you-go postpaid, usage packages, TPM-based billing), data management, model fine-tuning (covering Post-pretrain, SFT, and preference alignment, billed by training tokens, currently at a limited-time 5% discount), model distillation (knowledge distillation billed by fine-tuning instance duration at 112 yuan/hour, after a 5% discount), and model evaluation (judge models billed by inference). The platform also offers exclusive Skills such as Baidu Search, Baidu Baike, Baidu Maps, and document parsing, plus curated MCPs, as well as Baidu Security Scanning for Skill listing review and risk detection.
Developers who need to invoke LLMs via API, teams building agent applications, and technical professionals requiring enterprise-grade model services such as fine-tuning, distillation, and evaluation; the page also showcases industry application cases in government, enterprise, education, and finance.
Obtain a API Key on the 'Quick Start' page of the Qianfan console (Token Plan subscribers choose a dedicated API Key; other scenarios use the general API Key).
View billing methods, capability tags, context length, and input/output prices on model cards, select models as needed, and switch anytime.
Fields such as the selected model's model and API Key are auto-filled into the sample code—copy and run, with online debugging supported.
To use in a terminal or local AI Agent, install the Qianfan CLI, log in with one command, and invoke full-modal capabilities using natural language; you can also connect Qianfan Skills to Agents such as Comate and Claude Code via qianfan +connect.
For advanced capabilities, use modules such as data management, model fine-tuning, model distillation, and model evaluation to build your own models.
The API Key is the authentication credential for invoking model services; keep it safe and rotate it regularly. Token Plan subscriptions come in Personal and Enterprise editions, with each plan labeled 'Limited-time cancellation of 7-day quota'. The page mentions that some models are cheaper at night (22:00–8:00 the next day). Restrictions for other regions, accounts, or devices are not described on the current page; check the official website.
费用与套餐:Online inference supports pay-as-you-go postpaid, usage packages, and TPM-based billing; input/output are priced separately, cache hits save more, and some models are cheaper at night (22:00–8:00 the next day). Example prices: ERNIE-5.5-Flash-Preview input 0.001 yuan/thousand tokens (8:00-22:00), output 0.004 yuan/thousand tokens; DeepSeek-V4.1-Flash-1017 input 2 yuan/million tokens (daytime), 1 yuan/million tokens (nighttime). Token Plan subscriptions: Mini Personal 4.9 yuan/month (1,400 credits/month), Lite 19.9 yuan/month (11,500 credits/month), Standard 49.9 yuan/month (22,000+ credits/month, first month includes 1,760 bonus credits), Pro 99.9 yuan/month (45,000+ credits/month, first month includes 4,500 bonus credits), Max 299.9 yuan/month (165,000+ credits/month, first month includes 19,800 bonus credits). Model fine-tuning is currently at a limited-time 5% discount, and knowledge distillation is 112 yuan/hour (after a 5% discount). Prices may change; refer to the official website.