Not sure why the author thinks Anthropic's models spend more power and water than DeepSeek, there's no evidence of that. Their pricing has more to do with premium perception and less to do with COGS.

Everyone does optimization of model serving because it's good for every player in there.

(Also the water consumption thing is not a real issue.)

People aren't freaking out because self-hosting isn't a solve issue and it requires a lot of capital. The average company won't go and build an $1M GPU cluster just to self-host any model. If that gets commoditized then they'll start freaking out.

DeepSeek Flash 4.1 is generally what I use as as my workhorse. I use ChatGPT web to create the outline, then have Deep Seek built it out. Works generally pretty great.

Deepseek and many other models are heavily benchmaxxed. They aren't genuinely as good as the best frontier models.

Because the free api is mispriced, opus 5.5 on subscription is 10x cheaper. Also, the use case for flash models are

For coding, I rather spend 10x more than have even 1 bug but I'm only spending 2x 3x more if you count subscription cost.

This is a killer use case for something like customer support though.

Personally I have not used anything but 4.1 since it came out. I have a dataset that turns any model into pure hallucination machine, not only DeepSeek does not hallucinate, it builds new insights by combining its insights. It's not only cheap, its far better (at least for me)

we do. we moved 100% of our production traffic from Gemini to Deepseek. best decision we ever made.

Not only the model but the hardware it is running on. The Huawei chips they are using instead of NVIDIA are vastly less expensive. China is going to scale past the west. The idea that they are “only a few months behind” is today and many of us cannot even bring ourselves to admit it. The future is even more dramatic.

Mimo 2.6 Pro is even better.

I switched from DeepSeek 4.1 flash about 2 weeks ago for my Hermes sysadmin/coding agents and I am seeing better intelligence and lower overall spend.

https://artificialanalysis.ai/models/mimo-v2-6-pro

I’m also a heavy DS 4.1 Flash user—especially when it’s available at those off‑peak prices, which is an awesome deal. And, like you said, it’s genuinely powerful and very snappy. I’m planning to evaluate the differences between `reasoning_effort` settings today.