Anthropic’s Claude Haiku 4.5 in 6 Minutes

--- type: transcript date: 2025-10-15 youtube_id: jC_rX86O1Q8 --- # Transcript: Anthropic’s Claude Haiku 4.5 in 6 Minutes Anthropic has just released Claude Haiku 4.5, their latest small model that is available to all users today. Not only is this model cheaper, it's faster, but just 5 months ago when they released Claude Sonnet 4, this was the state-of-the-art model that was out there and now Claude Haiku 4.5 gives you similar levels of coding performance, but at a third of a cost and more than twice the speed. In this video, what I'm going to do is I'm going to quickly go over the announcement and then I'm going to show you a demonstration of what the model looks like for coding specific tasks. In terms of benchmarks, when we compare a Gentic coding task to GPD5, we have Haiku 4.5 at 73.3% on Sweetbench verified. Whereas with GPD5 on high mode, that ranks at 72.8% and GPD5 codeex respectively at 74.5%. On aentic terminal coding, when we compare that to GPD5, we have it ranking at 43.8% 8% whereas Gemini 2.5 Pro ranks at 25.3%. There is definitely very strong performance within computer use and they do specifically highlight that within the blog post mentioning that this model is very good at using certain applications including Claude for Chrome which was something they had just released just recently. In terms of some of the reviews, Claude Haiku 4.5 hit a sweet spot. We didn't think it was possible for nearfrontier coding quality with blazing speed and cost efficiency. In augment specific coding evaluation, it achieved 90% of set 4.5's performance matching much larger context window. We can see reviews from companies like Warp, Windsurf, Shopify, as well as a handful of others. Next up, in terms of some of the specifics of the model, it does support both text as well as images, and you will be able to pass in up to 200,000 tokens of context into the model. Now, in terms of availability, you'll be able to get this from the cloud API, from AWS, or from GCP, whichever you prefer. And in terms of pricing for the model, it is $1 per million tokens of input and $5 per million tokens of output. When we compare that to Sonnet 4.5, for instance, that is $3 per million tokens of input and $15 per million tokens of output. Now, in terms of how the price compares to some of the other competitors for the blended rate for some of the other models like GPD5 as well as Quen or Gemini Flash, we see that this model is more expensive than something like Gemini 2.5 Flash, but it does fall within the range of GPD5, GPT 2.5 Pro. Definitely is for a smaller model within the realm of the more premium models for some of the other providers, but especially for software development tasks as well as agentic coding. This is something that Anthropic is particularly good at and accelerating at. So it definitely could be a very interesting model for a lot of those types of use cases and I would imagine a lot of agentic coding tools or workflows would involve actually multiple series of different models. Next up in terms of accessing the models, you will be able to try this out completely for free on claude.ai. I'm going to do a couple demonstrations in terms of the model generation just to show you the speed as well as the capabilities. I'm going to say I want to create a beautiful SAS landing page. And as we can see here, it's going through and it's beginning to generate all of the different React as well as the JSX for what we had asked for. Now, additionally, you will be able to try out the model completely for free on the Claude web apps. I'll just demonstrate it here. I'm going to say I want to create a beautiful SAS landing page. Here we see the artifacts panel open up here and it begin to write out all of that relevant HTML for what we had asked for. One of the nice things with the web app is it does have this artifacts pane where you will be able to have these little interactive web apps just to give you a little bit of a sense in terms of its capabilities. And overall here, if I just take a quick look, here is what it has generated for us. We have these emojis. We also have this pricing section. And now just to demonstrate this within cloud code. The first thing that I'm going to do is I'm just going to create a blank Nex.js application within our directory here. And then once I have that, I'll go within our application. Once I have that, I can go ahead and run Claude. The first thing that you're going to have to do is if you forward slash, you're going to be able to type models. And within here, you can go ahead and click enter. And you can go down and you can select to default Cloud Haiku 4.5. And then you can go down and you can select the latest haiku model. I want to create a blog called developers digest that has a few example blog posts with beautiful syntax highlighting for some TypeScript tutorials. I'm going to go ahead and send that off. If you haven't used cloud code before, this is a really great demonstration in terms of some of those benchmarks like Swebench as well as Terminal Bench. What we'll see within here is we'll see the agent go through and execute all of the relevant commands to read through what we have within our project already. So, it's quickly going through. It's looking through the project. We can see that it's able to execute tools considerably faster than some of the other models. Having used cloud code for some time. The first thing that definitely stands out with the model is it does seem to execute everything much much faster than some of the other models like Opus as well as Sonnet 4.5. Within here, I can see it's going through, it's creating the blog post just like we had asked for. And the one thing with this type of task is it's a really good demonstration just to give you a sense of the agentic capabilities because it's both reading, it's writing, it's having to come up with the overall architecture and plan for how to actually execute and include this into our live working application. Here is what it has generated for us. On the homepage here, we can see we have the ability to view different articles. We have this hero here on the blog page. Now mind you, we didn't specify any styles. There are some nitpicks that I don't love like the emojis as well as the linear gradients that it included as a part of the initial generation. But if I take a look here and I go and I click through to view all the articles here, I see we have this other section here as well where a similar thing. We have some other emojis. We have a lot of that linear gradient within here as well. If I click through to one of the blog posts, I can actually see that it is functioning. And what's nice with this is it does have a nice little code block here equipped with proper syntax. Mind you, if I was to specify some of the things like the overall theme or style and look and feel, it could probably be that much more improved. But this was just from a sentence or two like I described. Kudos to the team at Anthropic for this latest release. And if you found this video useful, please like, comment, share, and subscribe. Otherwise, until the next