Experiments

The same task built with several tools, measured on cost and time with the same yardstick.

• Experiments

  Which AI Model for Work Automation? GPT-6 Luna vs Claude Haiku 5.5 on 11 Real Tasks

  11 real work tasks, from diagrams and slides to browser work, ran through five GPT and Claude models, three runs each, with scores and costs.

  October 8, 20268 min read

• Experiments

  GPT-6 Luna vs Claude Haiku 5.5: Results From Six Real Work Tasks

  I ran five AI models through six real work tasks, 90 runs total. Four tasks were perfect for all; most gaps came from self-introduction grading.

  October 8, 202611 min read

• Experiments

  Aside Browser: Should You Connect Claude or GPT? Accuracy, Speed, and Cost Test

  I ran three web tasks nine times each on Aside with Claude and GPT models. Three models scored perfectly; the cheapest, Luna, missed twice on shopping.

  October 6, 20268 min read

• Experiments

  Comparing 7 Ways to Build 5 Slides: The Most Expensive Scored Lowest

  I built the same 5 slides seven ways and measured cost, time, and spec compliance. The most expensive path followed the spec the least.

  September 9, 202612 min read