[mikesurowiec]

A rough idea of the price differences...

  Per 1k tokens        Input   |  Output
  Amazon Nova Micro: $0.000035 | $0.00014
  Amazon Nova Lite:  $0.00006  | $0.00024
  Amazon Nova Pro:   $0.0008   | $0.0032

  Claude 3.5 Sonnet: $0.003    | $0.015
  Claude 3.5 Haiku:  $0.0008   | $0.0004
  Claude 3 Opus:     $0.015    | $0.075

Source: AWS Bedrock Pricing https://aws.amazon.com/bedrock/pricing/

[sheepscreek]

It’s fascinating that Amazon is investing heavily in Anthropic while simultaneously competing with them.

[panabee]

Amazon is a retailer and strives to offer choice, whether of books or compute services.

AWS is the golden goose. If Amazon doesn't tie up Anthropic, AWS customers who need a SOTA LLM will spend on Azure or GCP.

Think of Anthropic as the "premium" brand -- say, the Duracell of LLMs.

Nova is Amazon's march toward a house brand, Amazon Basics if you will, that minimizes the need for Duracell and slashes cost for customers.

Not to mention the potential benefits of improving Alexa, which has inexcusably languished despite popularizing AI services.

:Edited for readability

[coredog64]

Minor nit: These days I think Ads has taken over as the golden goose, but that doesn’t diminish the contributions of AWS.

[jazzyjackson]

Is that why Amazon's product search is terrible? Because it's more profitable for them when I scroll through 5 pages of junk than if I can navigate immediately to the thing I want?

[wkat4242]

Yes because the sellers of that junk pay them to put it there. And most people give up and buy some of that. If it didn't work they wouldn't do it.

Cory Doctorow covered this specific phenomenon in his great article that coined the term enshittification.

[dotBen]

It’s fascinating that Amazon Web Services have so many overlapping and competing services to achieve the same objective. Efficiency/small footprint was never their approach :D

For example, look how many different types of database they offer (many achieve the same objective but different instantiation)

https://aws.amazon.com/products/?aws-products-all.sort-by=it...

[UltraSane]

Soon AWS is going to need an LLM just to recommend what service a customer should use.

[htrp]

Let me tell you about Amazon Q

[bushbaba]

To quote, “right tool for right job”.

[Jfurrrio2]

They are not competing those are offerings. "AWS has many offerings" is a completly different thing than saying they compete against each other.

[donavanm]

As others said the product isnt the model, its the API based token usage. Happily selling whatever model you need, with easy integrations from the rest of your aws stack, is the entire point.

[holub008]

Has anyone found TPM/RPM limits on Nova? Either they aren't limited, or the quotas haven't been published yet: https://docs.aws.amazon.com/general/latest/gr/bedrock.html#l...

[tmpz22]

Maybe they want to gauge demand for a bit first?

[Tepix]

I suggest you give the price per million token as seems to be the standard.

[8n4vidtmkvmk]

From my personal table https://i.imgur.com/WwL9XkG.png

Price is pretty good. I'm assuming 3.72 chars/tok on average though.. couldn't find that # anywhere.

[oblio]

I'm guessing they just copy pasted from the official docs page.

[warkdarrior]

Eyeballing it, Nova seems to be 1.5 order of magnitude cheaper than Claude, at all model sizes.

[Bilal_io]

You have added another zero for Haiku, its output cost is $0.004

[indigodaddy]

Thanks that had confused me when I compared same to Nova Pro

[mikesurowiec]

You're absolutely right, apologies!

[Havoc]

Doesn’t look particularly favourable versus deepseek and qwen. Main deepseek is about same price as smallest nova.

I guess it depends on how sensitive your data is

[jerrygoyal]

does anyone know performance benchmark