🚀 Agentic RAG Cost Calculator

Tính toán chi phí hệ thống AI với LangGraph - August 2025

📊 Bảng So sánh Model AI

Model	Provider	Context	Input $/M	Output $/M	Tool-use	Special Features
GLM-4.5	Zhipu AI	128K	$0.11	$0.28	Native Agent	Best value, MoE architecture
DeepSeek R1	DeepSeek	64K	$0.55	$2.19	Good	Fast reasoning
DeepSeek V3	DeepSeek	128K	$0.27	$1.09	Average	Cost-effective
Kimi K2	Moonshot	128K	$0.60	$2.50	Excellent	Agentic optimized
Qwen3-235B	Alibaba	256K	$0.735	$8.82	Good	Thinking mode
Gemini 2.5 Flash	Google	1M	$0.30	$2.50	Very Good	Massive context
Gemini 2.5 Pro	Google	1M	$1.25	$10.00	Excellent	Premium features
GPT-4.1	OpenAI	1M	$2.00	$8.00	Very Good	Industry standard
Claude Sonnet 4	Anthropic	200K	$3.00	$15.00	Excellent	Best for coding
Llama 3.3 70B	Meta	128K	$0.20	$0.20	Average	Open source

💰 Calculator Chi phí Hệ thống

Số lượng Users:

Queries/User/Day:

Working Days/Year:

Avg Input Tokens/Query:

Avg Output Tokens/Query:

Primary Model (70%):

Advanced Model (30%):

📋 Recommended Configurations

Configuration	Primary Model (70%)	Advanced Model (30%)	Monthly Cost	Annual Cost	Best For
Budget	GLM-4.5	Kimi K2	30M VND	360M VND	Startups, POCs
Balanced	Gemini 2.5 Flash	GLM-4.5	38M VND	457M VND	SMEs, Production
Premium	Gemini 2.5 Pro	Claude Sonnet 4	259M VND	3.1B VND	Enterprise, Critical

💡 Pro Tips:

GLM-4.5 có Native Agent support với giá cực tốt - ideal cho Agentic systems
Gemini 2.5 Flash cung cấp 1M context window - tốt cho document-heavy tasks
OpenRouter không markup giá và cung cấp auto-failover giữa providers
Batch processing có thể giảm 25-50% chi phí với nhiều providers
Test kỹ Chinese models trước khi deploy production

🔄 Token Usage Breakdown

Step	Input Tokens	Output Tokens	Frequency	Avg per Query
Query Analysis	500	100	100%	600
Query Rewriting	600	200	70%	560
RAG Retrieval	8,000	500	70%	5,950
Document Grading	8,500	200	70%	6,090
Web Search	2,000	1,000	20%	600
Answer Synthesis	9,000	800	100%	9,800
TOTAL	~15,000	~2,000		17,000