Claude 4 Sonnet Raises the Bar
october 1st: claude 4 sonnet went live.
if claude 4 opus was impressive, sonnet is impressive AND fast AND cheap.
the value proposition is getting absurd.
the positioning
- opus: maximum capability, premium use cases
- sonnet: great capability, practical speed and cost
sonnet is designed for everyday use. the model most people will actually interact with.
what's different from 3.5 sonnet
major improvements in:
- reasoning depth
- instruction following
- code generation
- reduced hallucinations
- nuanced responses
the gap between sonnet and opus is smaller than ever. which means sonnet is more useful than ever.
my usage
i switched my default personal use from opus to sonnet.
not because opus isn't good—it's the best we have. but for most tasks, sonnet is:
- fast enough
- smart enough
- much cheaper
i save opus for complex research and tough problems.
what this means for the market
the "best" model isn't always what people need. they need:
- good enough quality
- fast responses
- reasonable cost
- reliable availability
sonnet hits all of these. that's why it matters.
the competitive landscape
everyone's shipping models. gpt-5, gemini 2, various open-source options.
the field is crowded. which is good for users—competition drives improvement.
from inside anthropic: we're focused on our own work, not panicking about competitors. there's room for multiple good options.
insider perspective
watching the sonnet team ship this was cool.
the optimization work to make it fast. the eval refinement. the safety testing.
every release is a massive coordination effort.
looking ahead
opus 4.5 is coming? claude 5 eventually? things i can't discuss?
the pace doesn't slow down. each release is the new baseline.
used sonnet to help draft some code today. shipped faster because of it. the tools are good.