Model & agent platform
Qianfan MaaS
One OpenAI-compatible endpoint for every model family — ERNIE, DeepSeek, GLM, Qwen, Kimi and frontier models.
- 150+ SOTA models. Text, reasoning, vision, image and video, with ERNIE built in-house.
- Agents and tools. Model building, fine-tuning, MCP tools, knowledge bases and enterprise deployment.
- Performance edge. +25% output tokens per second, −16% time to first token versus the industry average.
- Operational controls. Elastic scaling, KV-cache optimization, sandbox execution and per-model billing.
curl · api.baiduqianfan.aiLive
curl https://api.baiduqianfan.ai/v1/chat/completions \ -H "Content-Type: application/json" \ -H "Authorization: Bearer bce-v3/ALTAK-…" \ -d '{ "model": "ernie-5.1", "messages": [ {"role": "user", "content": "Hello"} ], "stream": true }'
ernie-5.1128K context30%
deepseek-v4-pro1M context36%
glm-5.2500K context40%
kimi-k3256K context60%
+ 145 moreVision · video · speech · imageOne key
Base URL: https://api.baiduqianfan.ai/v1 · Bearer auth · default 60 RPM per key. Higher limits and reserved TPM are available through your solution architect.
150+
SOTA models, one catalogue
800K+
Developers on Qianfan
+25%
OTPS over industry average
−16%
Time to first token
Lowest ratesPayable share of published list — lower is better
Full rate card
ERNIE 5.130%Text & reasoning · 128K
DeepSeek V4 Pro / Flash33–36%Official upstream
GLM 5.240%500K context
Qwen 3.7 Plus60%Bilingual workloads
Kimi K360%Long context
Gemini / GPT / Claude65–78%Frontier models · official routed
Rates are the lowest payable share of the published international list price for standard capacity — lower is better. Volume, cache and reserved capacity set the final number; models marked on request are priced in the quotation.
Start
Not sure which line fits?
Tell us the workload — text, video, location or a full product — and we will scope the stack, the region and a firm rate.