Bluesky Thread

i’ve been sleeping. sonnet 3.7 is already out and available even on free ...

February 24, 2025 View original thread

i’ve been sleeping. sonnet 3.7 is **already** out and available even on free plans

www.anthropic.com/news/claude-...

A bar chart titled “Software engineering” with a subtitle “SWE-bench verified”, displaying accuracy percentages of different AI models. The y-axis represents accuracy, ranging from 0 to 80. The x-axis lists five AI models:
1. Claude 3.7 Sonnet – 62.3% accuracy, with an additional annotation showing 70.3% with custom scaffold.
2. Claude 3.5 Sonnet (new) – 49.0% accuracy.
3. OpenAI o1 – 48.9% accuracy.
4. OpenAI o3-mini (high) – 49.3% accuracy.
5. DeepSeek R1 – 49.2% accuracy.

The Claude 3.7 Sonnet model has the highest accuracy, significantly outperforming the others. The chart uses different shades for each bar, with Claude 3.7 Sonnet in a distinct reddish-orange, highlighting its superior performance.

Tim Kellogg @timkellogg.me

Anthropic leaked that they’re releasing Claude Sonnet 3.7

obvs very exciting

also hilarious that they had to skip 3.6 because they botched naming and called the last update “Claude Sonnet 3.5 (New)” so everyone called it 3.6

30 3

128k *output* tokens is interesting

who benefits?

1. writers (well, non-writers, ya know..)
2. C++ developers (who else writes files that long?)

8

More like this

Sonnet 4.5 is not fungible

Sonnet 4.5

Whoah.. sonnet was *not* distilled