Google announced a price reduction for Gemini Flash overnight. The announcement included a price cut of up to 85%.

I think I've written about five posts related to price cuts recently... It's safe to say that prices are being cut as much as possible.

A few days ago Price reduction related post, but a few days later... it was reduced again.

Here is an article about the new price reduction.

https://velog.io/@calmlake79/%EB%98%90..-%EA%B5%AC%EA%B8%80%EC%9D%B4-%EA%B0%80%EA%B2%A9%EC%9D%84

Here is the related Google document.

So, I've been posting a lot about prices lately, so I'm going to clean things up and stop posting about prices for a while.

OpenAI's Davinci, a product released at the beginning of last year, costs $2 per 1M tokens for input/output.

The price of Google's Gemini Flash, which was recently reduced, is $0.01875 / $0.075 for I/O. In terms of input, the price difference is a whopping 100 times. That means it's dropped to one-hundredth of the price. (And if you use Batch Prediction, this difference widens to 200 times.)

So, how about the performance? The difference is so big it's almost embarrassing to compare. Flash is a top-tier LLM in various fields, (excluding the flagship) and is inherently multilingual, while Davinci at the time was a product that was even ambiguous in Korean.

It's the first time I've seen an industry where price competition is happening so quickly.

Prices are almost cut in half every three months. When I started durumis last year, I was a little worried about translation costs, but even then, Google was cutting prices, so I thought, "Google will take care of it," and they actually did.

Now, at least when it comes to LLM, I don't have to worry about costs anymore, and I can add features without worrying.

There's no guarantee that prices will continue to fall, but even at the current level, I think I can do almost anything I want to do.

I hope you can look forward to a durumis that continues to improve.