Anthropic Just Dropped Claude Haiku 5.5 (Major Upgrade)

0:00 Anthropic just dropped their cheapest model in the Claude 5.5 family and it could change which model you use for a lot of your everyday work. It's called Haiku 5.5 and it comes right after the

0:10 release of Opus 5.5 and Sonnet 5.5 which have been extremely impressive. Now Haiku has always been the small and cheap model and honestly it hasn't been very capable in the past. So if this one

0:21 got the same kind of upgrade that Opus and Sonnet did then this can get really interesting. So, in this video, I'm going to break down exactly how cheap it is, how fast it is, and how capable it

0:30 is. I'll even give you examples of when you should use Haiku over Sauna or Opus, because a cheaper model doesn't always mean better. So, without further ado, let's dive right in. All right, so what

0:40 I'm going to do is walk through the article that Enthropic gave us and just give you the main talking points that matter to you. If you want to read it yourself, there will be a link in the

0:48 description to do so. Now, first things first. Enthropic is calling Haiku 5.5 the cheapest, fastest, and most capable small model they've ever released. If you're not familiar with the lineup,

0:58 Fable is their biggest model. Then we have Opus, then Saunet, and Haiku is the smallest and the cheapest. Anthropic says it's designed for high volume, cost-sensitive tasks. So, think quick,

1:08 repetitive work like summaries, sorting through emails, or pulling a number out of a report. They also say it's their fastest model to date, so it works really well for things like live

1:17 customer support and browser use. And the last Haiku model came out about a year ago, so this one has been a long time coming. Now, moving on to the biggest thing here, and that's the

1:27 price. Anthropic says that on average, Haiku 5.5 costs around 75% less than running Haiku 4.5. So, here is the breakdown side by side. For input tokens, the old Haiku was $1 per million

1:40 and Haiku 5.5 is 10. For output tokens, it went from $5 down to 50 cents. So if we do the math here, that is 90% lower. And if you compare it to Sonnet 5.5, that's $2 in and $10 out. So Haiku is 20

1:54 times cheaper than that one. Now, there is one catch, and it's this part right here. That 10-cent price only applies to prompts up to 100,000 tokens. Once you go over that, it's 50 in and $2.50 out,

2:07 which is still only half of what the old haiku cost. And to put that in perspective, 100,000 tokens is roughly 55,000 words. So that's about a short book in a single prompt. Anthropic says

2:18 about 90% of the requests to the old haiku were under that size. So most of the time you're getting the cheaper rate. Now, if we judge price alone, this opens up a lot of jobs that were

2:28 probably too expensive to automate before. Now, of course, cost is very important, but it doesn't really mean anything if the model isn't very capable. And like I said, the older

2:37 Haiku models have not been very impressive. So, this is the part that I personally was the most curious about. Here's exactly how it compares to Haiku 4.5. Openi small model and Sonnet 5.5.

2:48 And the thing that really stands out to me here is computer use. The old Haiku scored 15.7% on this test and Haiku 5.5 scored 72.4%. And moving on to knowledge work, which

3:00 is basically real life office tasks, it more than doubled the old haiku and it scored higher than openi small model as well. And on aentic coding, the old haik coup scored zero and haiku 5.5 is at

3:12 39.2%. Now, if you look over at sonnet 5.5, it's still ahead on every single one of those. And Enthropic says this right inside of the article. Sonnet and Opus

3:22 are obviously still the better choice for complex coding. One thing that I want to point out here is that this is the first Haiku model that comes with an adjustable effort setting, so you can

3:32 turn it up or down and decide if you want it cheaper or smarter. All right, so now that we've seen the price and the scores, let's talk about what you should be actually using Haiku for. Just

3:41 because it's the cheapest model doesn't necessarily mean you should switch everything over to it. Let's start at the top. Fable is for the hardest work like deep research or an agent that runs

3:49 for hours and hands you back a finished document or a slide deck. Opus is what I'd use for big complex projects like building out a full app or having Claude use your computer. Sonnet is the one

3:59 that you should be using every day. So things like writing content, analyzing your data, and most of your normal basic coding. And Haiku is for the jobs that are small, clear, and that you run a

4:10 lot. So think sorting every email that comes into your inbox, tagging support tickets, summarizing call transcripts, or pulling the totals off of a stack of invoices. It's obviously also great for

4:21 anything where speed matters, like a live support chat on your website. Now, where I wouldn't use it is anything with a lot of steps or anything where a mistake is going to be expensive. And if

4:30 we go back to the benchmarks, Haiku scored 39% on that coding test. And Sonnet scored 70%. So on the harder and more difficult tasks, that gap is pretty big. But based on the rest of the

4:41 scores, I think Haiku 5.5 is actually good enough to trust with simple jobs. And I really couldn't say this about the previous one. So here is the rule that I'd follow. If the job is simple and

4:50 you're running it over and over, start with Haiku, test it, and only move up to sonnet if it's getting things wrong. And if the job is complex, start with the biggest model. Now, the best part is is

4:59 that you don't have to just pick one of these. An Enthropic says Haiku pairs really well with Opus and Sonnet as a sub agent, which is basically a helper that the bigger model hands the smaller

5:09 jobs to. And one of the companies that had early access to test this out has a great example of this. They said while a bigger model builds a slide deck, Haiku goes into the financial report and pulls

5:19 the one number that the deck needs. This way, the expensive model only does the hard thinking and the cheap one does all of the leg work like scraping. Now, let's quickly talk about what some of

5:29 the early testers had to say. ASA said that they saw over a 30% reduction in latency compared to the model they use today. And latency is basically just how long you wait for a response. And Box

5:40 said that Haiku 5.5 scored 11 points higher than Haiku 4.5 at about half the latency. Now, Enthropic also put a couple of other updates at the very bottom of this article that I want to

5:50 talk about right now. The first is that Sonnet 5.5 just got cheaper. In a previous video that I did breaking down Sonnet 5.5 when it released, I told you that the cash reads on Sonnet 5.5 were

6:01 20 cents per million tokens. Well, they just cut that in half to 10 cents. And Enthropic says this makes Sonnet 5.5 around 20% cheaper on most Agentic work. And I think this is the part that most

6:12 people are going to miss. Enthropic is now giving Max and Team subscribers a monthly API credit. So, if you're on the Max 5X plan, you get $100 in credits every single month. If you're on max

6:24 20x, you get $200 and team plans get up to $500 that's shared across the team. Anyways, I hope you enjoyed this very quick breakdown. The AI race is heating up, so if you want to stay up to date as

6:34 a non techie, subscribe for more and I'll see you in the next video. Cheers.