What to know about Anthropic's new Claude 'Sonnet 5' AI model

Anthropic (ANTH.PVT) revealed its newest AI model, Claude Sonnet 5. Yahoo Finance Technology Editor Dan Howley explains what investors should know.

Video Transcript

00:00 Speaker A

Uh I ask you about Anthropic Dan this new AI model. Fill us in. What do we need to know?

00:04 Dan

Yeah, this is Sonnet 5. This is uh the third in I guess their, you know, five series uh of uh models. They obviously have Mythos 5, that's uh their huge model and then the Fable 5, that's kind of the model that's based on Mythos but has more security capabilities. Those are both still tied up with government regulation. Uh, the government has allowed some companies to get access to them, but that's that's a different story. This basically is supposed to be a a less expensive version uh of their models uh than something like their Opus 4.8 uh which was their prior generation uh highest end model.

01:03 Dan

And so really what they seem to be doing here is leaning into kind of reducing the cost per token uh when it comes to AI. Now tokens, they're basically a unit of measurement when it comes to AI. Uh a token can be anything from a short phrase to uh you know, pieces of a word. And so the idea here is that if uh they can cut down on costs, then companies will be less likely to want to move out of using AI as often. And we've seen some companies pull back on their AI spending. Meta uh has mentioned it, Uber uh uh blew through its own uh uh cap for AI spending. Amazon uh removing its leaderboard for how many uh tokens people are using. This whole kind of era of uh token maxing seems to be in the rear view mirror. And so uh companies like Genpic are responding with these models that are, you know, more efficient and uh uh cost less to use.

02:02 Speaker A

Which jobs, Dan could this new AI model impact? Like which profession should be paying very close attention here in your opinion?

02:12 Dan

I think generally it's it's an overall uh kind of, you know, normal model I guess you could nothing's really normal when it comes to AI, but I would say it's it's no particular job uh threat uh broadly. This is more of an agentic use case. So, uh that's really kind of where it's going to sit. Um, you know, it's it's not as though uh there's there's one that's uh uh better than the other when it comes to overall uh usage uh in different areas. This is more of a general model. So, yeah, it's it's going to be good for agentic uh and like I said, cheaper for that use case.

03:00 Speaker A

Is hallucination still an issue, Dan? I mean, can these AI agents, can they can they be trusted to work independently?

03:08 Dan

Yeah, I mean, there's always, you know, this this still kind of fear of hallucination for sure, right? I mean, as as good as these models get, that's going to be lingering in the background. You know, they they've made uh improvements uh to kind of cut down on those kind of issues, but you know, you're still going to get errors. I mean, you know, when it comes to just doing, you know, simple research at at points, I mean, uh my uh we were doing research to look up the dates for different earnings reports in the prior quarter, uh and Gemini was giving us the wrong dates entirely. Uh and so, you know, you get kind of things along those lines uh here and there. But I think it's gotten better. Um when it comes to AI, I really think that, you know, if if you are, you know, we're journalists obviously, so we have to double check our work, of course, every time. But I think that everybody who uses them should regardless.

View Comments