This document describes the configuration items and pricing standards of mainstream large language models (LLMs) recommended for use in Data Agent.
Disclaimer: FanRuan only provides recommended LLMs and connection methods for reference and takes no liability for any issues with the LLMs themselves.
As sensitive information, the API key can only be viewed and copied from the LLM platform during creation. You are advised to copy the API key immediately after creation to prevent it from being unavailable.
LLM
Configuration Item
Qwen
Endpoint (namely, base URL): https://dashscope.aliyuncs.com/compatible-mode/v1
API key: Register and log in to the official console, complete the token subscription,
and obtain the API key.
Common model names: qwen-turbo, qwen-plus, and qwen-max
DeepSeek
Endpoint (namely, base URL): https://api.deepseek.com/v1
API key: Register and log in to the DeepSeek official platform, go to the API keys
page, click Create new API key, and name a key for generation.
Common model names: deepseek-v4-flash and deepseek-v4-pro
Z.ai GLM
Endpoint (namely, base URL): https://api.z.ai/api/paas/v4/
API key: Register and log in to the Z.ai open platform, go to the API Keys page, and Create a new API key.
Common model names: glm-4-flash and glm-5.1
The same API key can be used for the related models compatible with the Azure protocol. The following lists creation steps for reference:
(1) Register and log in to the Azure portal.
(2) Search for Azure OpenAI.
(3) Create a resource. (The application submission and approval are required.)
(4) Choose Resource Management > Keys and Endpoint.
(5) Copy the API key.
For creation details, see the official documentation.
Azure OpenAI
(GPT series)
Endpoint (namely, base URL): https://{your resource name}.openai.azure.com
API version: 2025-01-01-preview
Common model names: gpt-4o, gpt-4o-mini, gpt-4.1, and o1/o3-mini
Anthropic Claude
(Azure-hosted)
Endpoint (namely, base URL): https://<your resource name>.cognitiveservices.azure.com/anthropic/v1
API version: not required for this model
Common model names: claude-opus-4.7, claude-sonnet-4.6, and
claude-haiku-4.5
Google Gemini
Endpoint (namely, base URL): https://<your resource name>.cognitiveservices.azure.com/google/v1
Common model names: gemini-3.1-pro and gemini-3.1-ultra
Recommended LLM
Input Price
(USD/Million Tokens)
Output Price
Description
qwen3.6-plus (full version)
$0.276 (tokens ≤ 256K)
$1.101 (tokens > 256K)
$1.651 (tokens ≤ 256K)
$6.602 (tokens > 256K)
· Chain-of-thought supported
· Billing instructions
qwen3-max-preview (full version)
$0.861 (tokens ≤ 32K)
$1.434 (32K < tokens ≤ 128K)
$2.151 (tokens > 128K)
$3.441 (tokens ≤ 32K)
$5.735 (32K<tokens ≤ 128K)
$8.602 (tokens > 128K)
· Discount available for context caching
kimi-k2.6 (full version)
$0.95/million tokens
$4/million tokens
· No token-based tiered pricing
glm-5.1 (full version)
Pay-as-you-go
MiniMax-M2.7 (full version)
$0.3/million tokens
$1.2/million tokens
· Only the non-chain-of-thought mode supported; Lowest cost
Single invocation cost = (Input token count/1,000,000 × Corresponding input unit price) + (Output token count/1,000,000 × Corresponding output unit price)
Assume that the average total per case is 20,000 tokens, of which:
Input: 18,000 tokens
Output: 2,000 tokens
Input-output ratio: 9:1
Billing based only on visible input/output tokens (chain-of-thought not taken into consideration)
Calculation example (using qwen3.6-plus as an example):
Cost = (18000/1000000 × 0.276) + (2000/1000000 × 1.651) = 0.036 + 0.024 = $0.00827/case
Cost Per Test Case (USD)
≈ 0.0078
≈ 0.00827
/
≈ 0.0251
≈ 0.02238