Thibault from OpenAI came for Opus. 🔫
He started by asking why companies still use Opus instead of Sol, and whether price really doesn't matter to them.
Then he explained his position in more detail.
"One OpenAI token doesn't equal one token from another model."
We compare AI prices in dollars per million tokens as if a token were a standardized unit like a gram or a kilowatt-hour. But it isn't.
Thibault gives a simple pizza example. One pizza is cut into 8 slices at $2 each, another into 16 slices at $1.25 each. In the second case, the slice is cheaper, but the whole pizza costs more.
The same applies to models. In a small test, GPT-5.6 Sol used 766 tokens for the same text that Claude Opus 5 needed 1,170 tokens for.
That's 34.5% fewer tokens for the same result.
So a lower price per million tokens doesn't necessarily mean the model is actually cheaper. What matters is the cost of a successful outcome, not the price of a single token.
Forwarded fromVibe Coding Daily
2August 17, 2026 188 2
