OpenAI GPT Version & Pricing (Update@202405)

計費方式

 * tokens 的 Input & Output 係分開計費的

 * 100 tokens is about 75 words.
    (1 token ~= 4 chars in English)

Context window

Max Message Count in Context

 * 128k = 128,000 tokens

Turbo

Turbo use less compute, thus its cheaper to run

 * 非 turbo 才能 follow complex instructions

ie.

GPT-3.5 Turbo

gpt-4-turbo@2024-04-09 (128k)

• $10 / $30 / 1M tokens

gpt4 (8k)

• $30.00 / $60 / 1M tokens

gpt4-32k

• $60.00 / $120 / 1M tokens

gpt-3.5-turbo

gpt-3.5-turbo Currently points to gpt-3.5-turbo-0125.

The last 4-digits number represent the date in MMDDMMDD

The snapshot that was taken for that version (與內容更新無關)

November 6, 2023 1 to name versions *-1106
January 25, 2024 4 to name versions *-0125

0125(16k)

• US$0.50/ US$1.50 / 1M tokens
• Training data: Up to Sep 2021

instruct(4k)

• US$1.50 / US$2.00 / 1M tokens

Fine-tuning models

gpt-3.5-turbo

# Training / Input / Output

US$8.00 / US$3.00 / US$6.00 / 1M tokens

System prompts

Guide ChatGPT’s responses in a more focused and controlled manner.

Prompt engineering

 - Your objective
 - Specific requirements

"token count" & "tokens used"

Token Count

It is a simple count of the individual words or tokens present in the text.

Tokens Used

refer to a specific type of tokens or keywords that have been utilized or triggered during the conversation.

所以 Used > Count