LLM World
The evidence-backed history of large language models.
Releases, families, labs, capabilities, context windows and pricing — every figure traced to the lab's own announcement, paper, model card or documentation. Where a lab publishes nothing, the record says so instead of guessing.
The history at a glance
How the context window grew, step by step →
The full timeline, month by month →
Why trust this record
-
Primary sources
Official announcements, papers, model cards and documentation. Reporting alone is never enough to mark a date verified.
The evidence rules -
Nothing is guessed
A figure no lab published stays empty. Undisclosed figures and unresearched ones are recorded as different things, because they are.
What is missing -
Claim-level evidence
Dates, context windows and parameter counts are traced to the specific source that states them, not to a source list.
Open any record -
Citations that cannot rot
Sources are cited by dated snapshot. A live page proves what it says today, not what it said when it was read.
How sources are cited -
Corrections in public
The record changes when the evidence changes, and every change is published — separating new research from figures this project had wrong. No other model tracker publishes its own corrections.
Read the changelog -
Built to be reused
The whole dataset is downloadable as JSON or CSV under CC BY 4.0, with an RSS feed of new releases.
JSON API
Prefer to browse? The timeline shows every release month by month, or start from the latest releases.