AI is moving too fast to watch all of it.
So I do the filtering.
Interesting models, strange demos, useful tools, and the occasional piece of internet chaos.
If it makes me stop scrolling, it belongs here.
3/3
The real metric isn’t price per token.
It’s cost per successful task.
How many tokens, retries, tool calls, and minutes did it take to get an answer you’d actually ship?
That’s the number that tells the real story.
1/3
AI pricing has a measurement problem.
“$ per million tokens” looks standardized. It isn’t.
Two models can handle the same job with different token counts, different output lengths, different numbers of retries, and different tool calls.
Cheap per token ≠ cheap AI.
2/3
Tokens are slices, not dinner.
One place charges $2 per slice and cuts the pizza into 8.
Another charges $1.25 per slice and cuts it into 16.
The second place advertises cheaper slices — but the whole pizza costs more.
Cheaper slices can still mean a more expensive pizza