Published: November 4, 2024
205
377
2.9k

Claude 3.5 Haiku is now available on our API, Amazon Bedrock, and Google Cloud's Vertex AI. Haiku is fast and particularly strong at coding. It outperforms state-of-the-art models—including GPT-4o—on SWE-bench Verified, which measures how models solve real software issues.

Image in tweet by Anthropic

During final testing, Haiku surpassed Claude 3 Opus, our previous flagship model, on many benchmarks—at a fraction of the cost. As a result, we've increased pricing for Claude 3.5 Haiku to reflect its increase in intelligence: http://anthropic.com/claude/ha...

Claude 3 Haiku remains available for use cases that benefit from image input or its lower price point. #model-names class="text-blue-500 hover:underline" target="_blank" rel="noopener noreferrer">https://docs.anthropic.com/en/...

@AnthropicAI As someone who loves your models and happily uses them daily, that last sentence is *not* going to go over well with people.

@AnthropicAI it's in the same level as 4o-mini in your benchmarks and it's 7x more expensive than 4o-mini, what kind of marketing is this?

Image in tweet by Anthropic
Image in tweet by Anthropic
Image in tweet by Anthropic

@AnthropicAI Increase pricing because it's smarter? lol

@AnthropicAI Huh? I had to re-read the tweet. It's your budget model that's competing against other budget models, why would you make it less competitive? People wanting a 'too cheap to meter' solution will now look elsewhere. Bad for the 'Haiku brand' and bad for your bottom line.

@AnthropicAI did you actually read what you just wrote? so you bumped up the price of the cheaper to run model because it's smarter on benchmarks? bad move

@AnthropicAI first paragraph: at a fraction of the cost second paragraph: we've increased pricing

@AnthropicAI will you also decrease the price if you release a dumb model or the next version underperforms vs openAI’s?

@AnthropicAI Did Haiku write that last sentence, too?

@AnthropicAI 1) what

@AnthropicAI 1) 3.5 haiku performs as good as 4o-mini on benchmarks 2) 3.5 haiku costs 7x more than 4o-mini no, thanks

@AnthropicAI Yikes..

@AnthropicAI Let Anthropic cook

@AnthropicAI AI S-Curve already flattening?

@AnthropicAI This should say "The model has no arbitrary guardrails and if you deploy with it, you should use an appropriate system prompt" Fuck off with the sAfEtY crap already

Image in tweet by Anthropic

@AnthropicAI this is a real shame.

@AnthropicAI Absolutely unhinged

@AnthropicAI What if one of your staff came to you and said hey I appreciate the job, I realize now that I’m here and see who else is working here I expect a higher salary. I am smarter than I expected to be. Please and thank you.

@AnthropicAI @sama opportunity striketh

@AnthropicAI Next OpenAI will announce they're decreasing price.

@AnthropicAI So, not a workhorse anymore. Weird pricing point considering Haiku 3 is not up to par with 4o-mini or Gemini 1.5 Flash. I'd say reconsider.

@AnthropicAI Someone in your marketing department needs to fired for thinking this was a good idea.

@AnthropicAI Does this mean that if someone finds the model performs worse than stated in your evals you'll issue eval-prorated refunds?

@AnthropicAI 4x the price is wild

@AnthropicAI Never thought I’d see “fraction of the cost” and “increased pricing” in the same tweet

@AnthropicAI Ummm... How did you write this publicly? This is such a strangely bad decision from y'all.

@AnthropicAI Increasing the price based solely on benchmark results is not something I would expect from @AnthropicAI. Shameful

@AnthropicAI What you'd expect: "As a result, we've decreased pricing for Claude 3 Opus to reflect its outdated intelligence" What we got: :(

@AnthropicAI come on - that's as much as Gemini Pro

@AnthropicAI Ahhhahahahaha, holy hell this is Epic! Why in the hell would you make an announcement like this?! Why so insanely expensive?! Why would anyone pay for that when gpt-4o-mini is leagues cheaper?! Whooweee, soneone had one helluva weekend over there

@AnthropicAI "increased pricing for Claude 3.5 Haiku to reflect its increase in intelligence"

@AnthropicAI this is trust breaching behavior developers will not like this

@AnthropicAI Dude the other mini models like GPT-4o-mini and Gemini 1.5 Flash are at 0.15$ and you’re over here at 1$ per 1m input tokens. Chances of me using haiku just went to 0.

@AnthropicAI Well, I was excited. Mood killer.

@AnthropicAI As someone who really appreciated anthropic until now… This last sentence isn‘t gonna go well.. (you know what i mean)

Share this thread

Read on Twitter

View original thread

Navigate thread

1/38