# ray-project/llm-numbers

Numbers every LLM developer should know

Repository: https://github.com/ray-project/llm-numbers
Canonical: https://ross.abutalabs.com/products/llm-numbers
License Family: other
Last push: 2024-01-16T10:15:43+00:00

## Health v2 (maintenance only)
Score: 29/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 0, release rhythm 35, longevity 86
- inputs: {"age_days": 1204, "days_push": 960, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 4315, forks 140 (observed 2026-08-28T04:08:41.819019+00:00)

## What it is
A reference document listing key numbers and rules of thumb that LLM developers should know for back-of-the-envelope calculations, inspired by Jeff Dean's 'Numbers every Engineer should know'. It covers tokenization ratios, prompt cost savings, and price comparisons between LLM models and services.

## Use cases
- estimate token counts from word counts for LLM billing
- compare costs between GPT-4 and GPT-3.5 Turbo for a task
- decide when to use embeddings lookup versus text generation
- learn how prompt phrasing like 'be concise' affects API costs
- do quick back-of-the-envelope math for LLM application budgets
- understand how context window sizes relate to document length

## When to choose
- you are new to LLM development and need intuition for token-based pricing
- you want quick reference numbers for estimating LLM application costs
- you are deciding between model tiers for one-off versus in-cycle tasks
- you need a shared vocabulary of LLM cost facts for your team

## When to avoid
- you need up-to-date pricing since the document was last updated in 2023
- you want runnable code or tooling rather than a reference document
- you need vendor-neutral numbers beyond OpenAI, Anthropic, and Cohere examples
- you require rigorous benchmarks rather than rules of thumb

## Facets
- artifact type: learning-resource
- maturity: maintenance
- function: llm-inference, prompt-engineering, developer-tools
- domain: large-language-models, tutorials, developer-tools
- platform: cross-platform
- tags: reference-guide, llm-cost-optimization, back-of-envelope-calculations, token-pricing, documentation

## Member repositories
- ray-project/llm-numbers (main) score 29

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:41.819019+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:21:46.630859+00:00, confidence not recorded.
  - readme: https://github.com/ray-project/llm-numbers (fetched 2026-08-28T04:08:41.819019+00:00, sha 361ca1c852d0)
- Data as of 2026-08-30T08:39:29.467469+00:00.