Input Tokens
The pieces of text, like words or parts of words, that an AI reads to understand a request.
Input tokens are the basic units of text that AI models like ChatGPT or Gemini process. Think of them as the AI's alphabet. A single word might be one token, or it could be broken into smaller pieces, especially if it's a complex or uncommon word.
When you type a question or a prompt into an AI, it first breaks that text down into these tokens. The AI then uses these tokens to understand your request and generate a response. The number of tokens in your input, and the AI's output, often affects how much it costs to use the AI and how long it takes to get an answer.
Input tokens are discrete units of text, typically words or sub-word segments, that an LLM processes to understand a given prompt or query. These tokens are generated by a tokenizer, a component that converts raw text into a numerical representation the model can ingest. The efficiency and effectiveness of an LLM are often constrained by its context window, which defines the maximum number of input and output tokens it can handle simultaneously.
Recent discussions highlight how the length of input tokens affects AI performance, with some models struggling as context windows grow. Companies are also adjusting pricing and limits based on input token usage, making it a key factor in cost and capability.
- Better prompt caching for GPT-6OpenAI · OpenAI News ·
- GPT-5.6 Cyber ModelOpenAI · Model Documentation ·
- GPT-6 Astra ModelOpenAI · Model Documentation ·
- Context Rot: How Increasing Input Tokens Impacts LLM Performance (Paper Analysis)Anthropic · Yannic Kilcher ·
- Anthropic Claude
- Anthropic's models process input tokens as part of their text-based interactions.
- Google Gemini
- Google Gemini's research explores how increasing input tokens can impact LLM performance, noting potential degradation with longer contexts.
- OpenAI ChatGPT
- OpenAI's models, like GPT-6, are being optimized with features like prompt caching that consider input token efficiency, and new models have specific token limits.
- Words
- While words are often tokens, complex words can be split into multiple to
- Characters
- Tokens are generally larger than individual characters, representing
- Sentences
- A sentence is composed of many tokens; tokens are the smallest units
The concept of tokenization, breaking text into smaller units for processing, has roots in natural language processing research. Specific tokenization methods like Byte Pair Encoding (BPE) were popularized by models like GPT-2. Source ↗
Open ChatGPT or Google Gemini and paste a long article into the prompt box. Then, ask the AI to summarize it. Notice how quickly it responds and if it mentions any limitations related to the length of your request.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- Cost tipMaximize GPT-6 Cache Discounts with 30-Minute WindowsReuse shared prefixes within 30 minutes for significant discounts on cached input tokens. · 5 min
- Cost tipUnderstand GPT-6 Astra cache write pricingCache writes for GPT-6 Astra are billed at 1.25x the uncached input token rate. · 1 min
- PromptRole-play a student struggling with math for AI tutor simulationSimulate a student's math struggles and uncertainty for AI tutor evaluation. · 5 min
- TemplateReusable workflow templates for healthcare teamsTeams can create and share templates for common workflows like documentation or prior authorizations. · 15 min
- TemplateTemplate for building an AI tool tailored to your workA structured approach to defining and building a custom AI tool for specific business needs. · 15 min
Questions people ask
- How many tokens is a word?
- It varies. A common word like 'cat' might be one token. A longer word like 'tokenization' might be split into multiple tokens like 'token' and 'ization'.
- Does the AI count spaces as tokens?
- Sometimes. Spaces are usually handled as part of a token or as a separate token depending on the specific tokenizer used by the AI.
First seen on Subvolts Jul 23, 2025. Updated Sep 29, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.