# Tokens > Tokens are the small chunks of text that AI language models read and write, roughly three-quarters of a word each. They determine both the cost of a request and how much of a construction document fits in a model's context window. - Stages: Preconstruction, Estimating - Concepts: LLMs - Published: 2026-08-28 - Canonical: https://buildr.com/library/tokens ## Why it matters in construction Tokens are the unit providers use to measure AI requests. When a vendor says a tool can handle "documents up to 200 pages," it is translating a token limit into pages. When a usage-based contract quotes cost per million tokens, that is what you are buying. When a tool truncates a long addendum and misses a scope change, token limits may be the reason. Precon does not need to count tokens by hand. A rough exchange rate between pages and tokens helps the team read a vendor's pricing sheet and recognize when a document may be too big to send whole. ## How it works 1. **Tokenization.** Before a model sees any text, a tokenizer splits it into pieces. Common words are one token; rarer words break into several. "Drywall" might be one token; "firestopping" might be two or three. Numbers and part codes are often several tokens each. 2. **Rule of thumb.** In English, one token is about four characters or three-quarters of a word. A thousand tokens is about 750 words, though page counts vary with formatting. 3. **Input and output.** Everything sent to the model counts as input tokens, and everything it writes back counts as output tokens. Output tokens usually cost several times more. 4. **Cost.** Providers bill per million tokens. The cost to read a 40-page sub proposal and return a one-page leveled summary varies by model and pricing agreement. 5. **Limits.** The context window is expressed in tokens, so token count decides whether a document fits. Tables, scanned PDFs run through OCR, and drawings with heavy annotation produce far more tokens per page than clean prose. That is one reason plan sets are a poor fit for feeding to a model whole. ## Example in practice Suppose a GC is comparing two AI proposal-review tools. Vendor A prices at a flat $400 per seat per month. Vendor B charges $25 per million tokens processed. The estimating team pulls last year's numbers: about 1,800 sub proposals received, averaging 12 pages, roughly 8,000 tokens each, plus about 1,500 tokens of output per proposal. That is about 17 million tokens a year, or around $430 at Vendor B's rate for the whole team. Vendor A's per-seat pricing across five estimators would be $24,000. The token math does not decide the purchase, but it gives the team a clearer way to compare price against what the product can do. ## Go deeper - [GPT Models for Contractors: A Plain-English Guide](/blog/gpt-models-for-contractors.md) - [Best AI Preconstruction Software for GCs 2026](/blog/best-ai-preconstruction-software.md) ## Related terms - [Context Window](/library/context-window.md): A context window is the amount of text, measured in tokens, that an AI language model can consider in one request. It limits how much of a construction spec book or proposal set the model can see at once. - [Large Language Model (LLM)](/library/large-language-model.md): A large language model (LLM) is an AI system trained on large amounts of text that can read, summarize, and generate language. In construction, it can help parse RFPs, sub proposals, and specifications for preconstruction teams. - [Retrieval-Augmented Generation (RAG)](/library/retrieval-augmented-generation.md): Retrieval-Augmented Generation (RAG) finds relevant pages in a company's own documents and gives them to an AI language model before it answers. It grounds a construction team's responses in its actual specs, proposals, and project history instead of the model's general knowledge. - [AI Vendor Evaluation](/library/ai-vendor-evaluation.md): AI vendor evaluation is how a construction company assesses AI software before buying it: data handling, security, accuracy on real bid documents, integration with existing systems, and whether AI is central to the product or added on. ## Referenced by - [Embeddings and Semantic Search](/library/embeddings-and-semantic-search.md): Embeddings are numeric representations of text that place similar ideas near each other even when the wording differs. Semantic search uses them to find construction documents, subs, or past projects by meaning rather than exact keywords, so acoustical ceiling can match ACT and lay-in tile. - [Reasoning Models](/library/reasoning-models.md): Reasoning models are AI language models designed to spend more effort on a problem before answering. They can be useful for multi-step construction judgments, such as reconciling a spec conflict across three divisions or scoring a go/no-go with competing criteria. ## FAQ ### How many tokens are in a page of a spec book? A dense spec page is usually 500 to 800 tokens. A full project manual of 800 pages lands somewhere around 500,000 tokens, though tables and scanned pages vary. ### Why should a precon manager care about tokens? Because tokens are how AI usage is priced and limited. If a vendor charges by usage or caps documents by size, tokens are the number behind that. They also explain why some documents get truncated. ### Are tokens the same across every model? No. Each model family uses its own tokenizer, so the same proposal might be 9,000 tokens in one model and 10,500 in another. The ratios are close enough for planning.