Anthropic Launches Claude 3.5 Sonnet
Key Points
- Anthropic released an upgraded Claude 3.5 Sonnet—keeping the same name but delivering substantially better performance, especially on coding evaluations.
- They also launched a faster Haiku model that matches the quality of the older Opus version, indicating a shift toward consistent naming conventions.
- Anthropic’s API business, though smaller in absolute terms than OpenAI’s, is growing about five‑times year‑over‑year and is on track to surpass OpenAI’s API revenue.
- The new release emphasizes developer tooling, including a beta feature that lets Claude Sonnet control a computer screen for agentic workflows such as automated user‑acceptance testing in Replit’s “repet” agent.
Full Transcript
# Anthropic Launches Claude 3.5 Sonnet **Source:** [https://www.youtube.com/watch?v=GMyJSd2yJYA](https://www.youtube.com/watch?v=GMyJSd2yJYA) **Duration:** 00:03:00 ## Summary - Anthropic released an upgraded Claude 3.5 Sonnet—keeping the same name but delivering substantially better performance, especially on coding evaluations. - They also launched a faster Haiku model that matches the quality of the older Opus version, indicating a shift toward consistent naming conventions. - Anthropic’s API business, though smaller in absolute terms than OpenAI’s, is growing about five‑times year‑over‑year and is on track to surpass OpenAI’s API revenue. - The new release emphasizes developer tooling, including a beta feature that lets Claude Sonnet control a computer screen for agentic workflows such as automated user‑acceptance testing in Replit’s “repet” agent. ## Sections - [00:00:00](https://www.youtube.com/watch?v=GMyJSd2yJYA&t=0s) **Anthropic's Strategic Claude 3.5 Release** - Anthropic unveiled an upgraded Claude 3.5 Sonnet—keeping the familiar name but delivering markedly better evaluation scores and coding ability—to signal a market‑share play against OpenAI while emphasizing its rapidly growing API business. ## Full Transcript
anthropic dropped a huge release I want
to get into it and I want to tell you
kind of how it shapes our understanding
of the players in this field because
anthropic and open AI are starting to
take up different niches and we need to
understand that so first what did they
drop they dropped an upgraded Claude 3.5
Sonet they did not rename it they did
not give it a new number again these
naming conventions are bizarre it's 3.5
sonnet just like you've always known and
loved except it's a lot better like it's
substantially better on the evals it's
way better at coding which was always
good at and that's going to come back
around you'll see why in a second and so
they dropped this new thing they give it
the same name for reasons I don't fully
understand they then drop an upgraded
version of haiku which they say is as
good as their old version of Opus but
much much faster we are starting to see
that anthropic has picked naming
conventions and are going to stick to
them and so I
think this may be a bit of a signal to
the other major players in the field to
gole gole and to open AI that anthropic
has more in the tank that they are not
releasing a new named model deliberately
and that this should be viewed as a
waiting game release to continue to gain
market share where they play best and
that gets to the strategy here so
anthropic is about a quarter of the size
of open AI but their API business is
much stronger their API business is
growing much faster than open ai's API
business and is on Pace to overtake it
so year-over-year open AI uh open AI is
doubling its API business and they're
over a billion dollars well anthropic is
at somewhere around a $700 million run
rate but they're growing 5x
year-over-year like they're growing much
much faster on that API side and that is
why in their press release they
highlighted the dev tooling capacities
they made Sonet better for coders they
made sure that they called out how Dev
tooling is affected by their new
experimental computer screen usage and
this is this is the big thing right they
are going to let Claude Sonet control
your computer screen it's the first hint
we've seen of where these agentic
workflows are going to go it's still in
public beta it's not fully released yet
but people can access it and use it and
the use case they call out is developer
friendly they are saying that repet is
using this in repet agent which is their
AI Builder to help user acceptance
testing for things that users are
building so that the repet agent
actually goes and like tests it on a
screen the way you would
um and improves the quality of the code
that way so they're doubling down on
their API positioning I think that's
smart IT differentiates them from open
AI they are not changing the name for
reasons I don't fully understand but it
is a big release uh it's substantially
better on the coding evals we'll have to
see how it looks in the wild uh but I'm
excited to play with Sonet 3.5 the new
version uh new and improved right so let
me know what you're doing let me know
how you're coding cheers