LLMCost

RAG Pipeline LLM Cost

Retrieval-augmented generation over your documents. Monthly estimates at 30,000 requests/mo, avg 4,000 input / 500 output tokens per request.

#ModelEst. monthly costvs cheapest
1Qwen: Qwen3.7 Flash$5.55—
2DeepSeek: DeepSeek V4.1 Flash (batch)$8.251.5×
3Anthropic: Claude Haiku 5.5 (batch)$9.751.8×
4OpenAI: GPT-6 Luna Pro (batch)$9.751.8×
5OpenAI: GPT-6 Luna (batch)$9.751.8×
6DeepSeek: DeepSeek V4.1 Flash$11.512.1×
7OpenAI: GPT-5 Nano$12.002.2×
8Google: Gemini 2.5 Flash Lite$18.003.2×
9Anthropic: Claude Haiku 5.5$19.503.5×
10OpenAI: GPT-6 Luna Pro$19.503.5×
11OpenAI: GPT-6 Luna$19.503.5×
12Qwen: Qwen3.8 Omni Flash$25.054.5×
13Qwen: Qwen3.8 Flash$25.054.5×
14Qwen: Qwen3.8 27B$25.054.5×
15DeepSeek: DeepSeek V4.1 Flash 1017$27.004.9×
16Meta: Llama 4 Maverick$32.295.8×
17DeepSeek: DeepSeek V4 Flash Vision Exp$35.576.4×
18OpenAI: GPT-5.6 Luna$42.007.6×
19OpenAI: GPT-5.4 Nano$42.757.7×
20DeepSeek: DeepSeek V3.1$44.258.0×
21OpenAI: GPT-5 Mini$60.0010.8×
22DeepSeek: DeepSeek V4 Flash 0731$72.6013.1×
23Google: Gemini 3.8 Flash (batch)$73.1313.2×
24Google: Gemini 3.7 Flash (batch)$73.1313.2×
25Google: Gemini 3.5 Flash Lite$73.5013.2×
26Google: Gemini 2.5 Flash$73.5013.2×
27DeepSeek: DeepSeek V4 Pro 0813$108.9019.6×
28Amazon: Nova Pro 1.0$144.0025.9×
29Google: Gemini 3.7 Flash$146.2526.4×
30Google: Gemini 3.6 Flash$146.2526.4×
31OpenAI: GPT-5.4 Mini$157.5028.4×
32Google: Gemini 3.8 Flash$160.8829.0×
33Google: Gemini 4 Argon (batch)$195.0035.1×
34OpenAI: GPT-6.1 Sol Pro (batch)$195.0035.1×
35OpenAI: GPT-6.1 Sol (batch)$195.0035.1×
36Anthropic: Claude Sonnet 5.5 (batch)$195.0035.1×
37OpenAI: GPT-6 Sol Pro (batch)$195.0035.1×
38OpenAI: GPT-6 Sol (batch)$195.0035.1×
39OpenAI: o4 Mini$198.0035.7×
40Mistral: Mistral Large 4$225.9040.7×
41Google: Nano Banana 2.1$292.5052.7×
42OpenAI: GPT-5$300.0054.1×
43Google: Gemini 2.5 Pro$300.0054.1×
44Google: Gemini 3.5 Flash$315.0056.8×
45xAI: Grok 4.7$330.0059.5×
46Qwen: Qwen3.8 Max (0902)$330.0059.5×
47Qwen: Qwen3.8 2.4T A95B$330.0059.5×
48xAI: Grok 4.6$330.0059.5×
49OpenAI: o3$360.0064.9×
50Anthropic: Claude Sonnet 5$390.0070.3×
51Google: Gemini 4 Argon$390.0070.3×
52OpenAI: GPT-6.1 Sol Pro$390.0070.3×
53OpenAI: GPT-6.1 Sol$390.0070.3×
54Anthropic: Claude Sonnet 5.5$390.0070.3×
55OpenAI: GPT-6 Sol Pro$390.0070.3×
56OpenAI: GPT-6 Sol$390.0070.3×
57Anthropic: Claude Opus 5.5 (batch)$390.0070.3×
58OpenAI: GPT-5.6 Terra Pro$420.0075.7×
59Anthropic: Claude Opus 5 (batch)$487.5087.8×
60OpenAI: GPT-5.4$525.0094.6×
61Anthropic: Claude Sonnet 4.5$585.00105.4×
62Qwen: Qwen3.8 Max Prime$660.00118.9×
63OpenAI: GPT-5.6 Sol$780.00140.5×
64Anthropic: Claude Opus 5.5$780.00140.5×
65Anthropic: Claude Opus 5$975.00175.7×
66OpenAI: GPT-6 Astra (batch)$975.00175.7×
67OpenAI: GPT-6 Astra Pro (batch)$975.00175.7×
68Anthropic: Claude Fable 5.1 (batch)$975.00175.7×
69OpenAI: GPT-5.5$1050.00189.2×
70Anthropic: Claude Fable 5$1950.00351.4×
71OpenAI: GPT-6 Astra$1950.00351.4×
72OpenAI: GPT-6 Astra Pro$1950.00351.4×
73Anthropic: Claude Fable 5.1$1950.00351.4×
74Anthropic: Claude Opus 4.1$2925.00527.0×
75OpenAI: GPT-5 Pro$3600.00648.6×
76OpenAI: GPT-5.5 Pro$6300.001135.1×

Cost-saving tips for rag pipelines

  • High request volume × large input = budget models shine. Quality delta is often negligible.
  • Deduplicate retrieved chunks before sending to the model.
  • Consider smaller chunks + top-3 retrieval instead of top-10 to cut input tokens by half.

Customize the numbers with the interactive calculator · All scenarios