4 ms·

In a single request? Oof. I was benchmarking some models the other day via openrouter and I got the distinct impression some of these models treat the thinking

In a single request? Oof. I was benchmarking some models the other day via openrouter and I got the distinct impression some of these models treat the thinking token budget as a target rather than a maximum.

stavros 1y ago
No no, overall, trying things a few times.