Potrošnja tokena generativnog AI

Ili ti koliko nas sve to košta? Potrošnja po modelima.

Svaki AI model ima različite mogućnosti i cene. Cene za AI modele se obračunavaju na osnovu korišćenih tokena. Tokeni su delovi reči, gdje je 1000 tokena oko 750 reči. Ovaj paragraf ima oko 80 tokena.

LLM Inference

OpenAI	Prompt	Response
GPT-4	$36.00/ 1M tokens	$72.00/ 1M tokens
GPT-4 Turbo	$12.50/ 1M tokens	$37.50/ 1M tokens
GPT-3.5 Instruct	$1.95/ 1M tokens	$2.60/ 1M tokens
GPT-3.5	$1.30/ 1M tokens	$2.60/ 1M tokens

Anthropic	Prompt	Response
Claude 3 Opus	$18.00/ 1M tokens	$90.00/ 1M tokens
Claude 2	$10.00/ 1M tokens	$30.00/ 1M tokens
Claude 3 Sonnet	$3.75/ 1M tokens	$18.75/ 1M tokens
Claude Instant	$1.04/ 1M tokens	$3.12/ 1M tokens
Claude 3 Haiku	$0.33/ 1M tokens	$1.63/ 1M tokens

Google	Prompt	Response
PaLM 2	$0.33/ 1M characters	$0.65/ 1M characters
Gemini Pro	$0.17/ 1M characters	$0.49/ 1M characters
Gemini Pro Vision	$0.17/ 1M characters	$0.49/ 1M characters

Mistral AI	Prompt	Response
Mixtral 8x7B Instruct	$0.36/ 1M tokens	$0.36/ 1M tokens
Mistral 7B Instruct	$0.17/ 1M tokens	$0.17/ 1M tokens

Meta	Prompt	Response
Llama-2 70B Chat	$0.91/ 1M tokens	$1.17/ 1M tokens
Code Llama	$0.78/ 1M tokens	$0.78/ 1M tokens
Llama-2 13B Chat	$0.29/ 1M tokens	$0.29/ 1M tokens

Data Sources – baza podataka

Data Vectorization		$0.13/ 1M tokens
Data Query		$0.65/ 1K calls

Function Calls – poziv funkcije

Scrape URL		$0.43/ 1K calls
Google Web Search		$1.25/ 1K calls