Data: https://
artificialanalysis.ai/leaderboards/m
odels
…
Get the evident-charts skill: https://
github.com/rhiever/eviden
t-charts
… October 6 snapshot. Costs are per benchmark task; similar aggregate scores don't guarantee equal performance on your tasks.
@randal_olson
-
AI model leaderboard snapshot comparing benchmark costs
By
–
-
Reasoning effort cost tradeoffs between Opus and Sonnet models
By
–
Max reasoning effort can nearly double cost for minimal Intelligence Index gains. Before switching to max, compare models. Opus 5.5 at xhigh roughly matches Sonnet 5.5 at max for less than half the cost. Made with evident-charts, my open source skill for explanatory charts.
-
AI coding agents and TDD best practices via plugin
By
–
Gotcha. I use @obra
's superpowers plugin, which explicitly guides AI coding agents to follow TDD best practices, so I haven't tested if the models on their own know to do that. If I had to bet though, they probably do, but maybe they'd follow it inconsistently. -
Managing AI agents to avoid unit test bloat
By
–
Of course, you need to apply pressure on the AI agents to not write too many unit tests, otherwise you get extreme unit test bloat.
-
Guardrails and signals for AI coding agents
By
–
AI coding agents work best when they have guardrails to keep their code sane and a signal to measure improvement against. Unit tests, linting, etc are cheap signals to keep the AI coding agent in check. Also side benefit of producing a unit test suite that catches regressions.
-
Smartest AI model you can afford at the speed you need
By
–
Question 3 from @ArtificialAnlys data: what's the smartest AI model you can afford at the speed you need? Past about a minute, waiting longer barely helps. Paying for a better model does. Made with evident-charts, my open source agent skill for explanatory charts.
-
Deep-research skill automatically fires on simple queries, wasting tokens
By
–
I was excited to see the /deep-research skill ship in Claude Code until it started firing automatically on simple queries. Whoever built this skill clearly doesn't have token limits – holy moly!
-
AI house staging for dollars, solves window misplacement problem
By
–
I just staged a house with AI for a few dollars. Staging used to run a couple thousand a listing, so this still feels a little wild. Problem is, AI models sometimes redraw the room while they furnish it, and the real windows and doors land on the wrong walls. So I built a
-
Modify a brand asset with AI: prompt and pray
By
–
Modifying a brand asset with an AI image model is often a matter of 'prompt and pray'. Each result seems almost done, so you regenerate and hope it will get it right next time. Goodeye checks each modification and locks your
-
AI job impact hits youngest workers, down 12%
By
–
Since ChatGPT, employment for workers aged 22 to 25 in the most AI-exposed jobs is down 12%. Every older age group held steady. Why only the youngest? Source: https://
randalolson.com/2026/06/22/ai-
jobs-hit-youngest-workers/?utm_source=x&utm_medium=social&utm_campaign=ai-jobs-hit-youngest-workers
…