Locul / Research

Open data · CC BY 4.0

Research on how AI actually works at work

Studies we ran ourselves, with the method and the raw data published in full. Download it, check it, reuse it with a link back.

• 
  Benchmark · 24 Oct 2026

  Office AI Benchmark 2026: Which office jobs can a laptop AI model do?

  50 real office tasks, from dictated email replies to Excel formulas checked on real spreadsheets, run through 14 cloud and laptop models against a fixed pass bar. A laptop-size model got 18 of 21 assistive email tasks done; long, messy threads still need a frontier model. Smallest model per task, cost and speed.

  Read the study →
• 
  Original study · 8 Oct 2026

  The State of CLAUDE.md

  What thousands of public CLAUDE.md files on GitHub actually contain: length, sections, commands, imports and how often anyone edits them. With a template built from the data and a checker.

  Read the study →
• 
  Reference table · 7 Oct 2026

  AI memory and context limits by plan

  What ChatGPT, Claude, Gemini, Copilot and Perplexity remember on free, paid and team plans, from context windows to retention. Every cell links to the vendor's own documentation.

  Read the study →