Interesting...
Sonnet 5 isn't made for coding,
It's made for knowledge work.
On CursorBench it's roughly as expensive as Opus 4.8, but worse. It's also more expensive than GPT 5.5 and worse.
However, on benchmarks like GDPval and Real-World Finance v2 that test knowledge work capabilities, Sonnet 5 beats Opus 4.8 and is cheaper.
Coding agents are table stakes now. The real race is everything else.
Trying Sonnet 5 inside Strawberry now, it's good! 🔥