this post was submitted on 08 Jul 2026
69 points (100.0% liked)

technology

24418 readers
156 users here now

On the road to fully automated luxury gay space communism.

Spreading Linux propaganda since 2020

Rules:

founded 6 years ago
MODERATORS
 

Microsoft is replacing OpenAI and Anthropic models with its own MAI models in Excel and Outlook to cut soaring AI costs.

Heavy AI users can cost up to $14,000 on a $200 subscription, forcing companies to subsidize usage and seek cheaper alternatives.

Microsoft's in-house models perform significantly worse than premium rivals, closer to open-source options like DeepSeek than Anthropic's best models.

Businesses are switching to Chinese AI providers charging 18 cents per million tokens versus $4 for Western models, reshaping the industry.

top 17 comments
sorted by: hot top controversial new old
[–] BeanisBrain@hexbear.net 54 points 1 week ago (1 children)

Businesses are switching to Chinese AI providers charging 18 cents per million tokens versus $4 for Western models, reshaping the industry.

lmao Western AI companies are fucked. You can't compete with someone who's offering comparable service for 5% of your price.

[–] Runcible@hexbear.net 28 points 1 week ago* (last edited 1 week ago) (1 children)

models perform significantly worse

What a frustratingly unqualified statement. I can totally admit that I don't know how to correctly interpret this but the comparisons I saw showed DeepSeek as only a few percentage points off the Anthropic benchmarks at the time, could be a big deal, could be imperceptible, I have no idea.

edit: quote is from summary, not the comment

[–] Tabitha@hexbear.net 17 points 1 week ago

yeah like OpenAI/Claude has superior tech but DeepSeek is probably already overkill for 99+% of real world LLM calls.

[–] FnordPrefect@hexbear.net 36 points 1 week ago
[–] InevitableSwing@hexbear.net 34 points 1 week ago (2 children)

Microsoft's in-house models perform significantly worse than...

I laughed out loud at that. I assumed so but it's still really funny to see it written out. It's basically a meme format...

Microsoft's _________________________ perform(s) significantly worse than...

[–] DogThatWentGorp@hexbear.net 12 points 1 week ago (1 children)

If it's anything like their other software having the most obscure and Byzantine menus and features here's how it'll go:

"Michael's Slop AI I want to write a document"

"Okay. Opening up document center page."

"No just open word"

"Hi. : ). The document center page has everything you need to get creating! Your subscription has given you access to a TON of new features you can browse at any time"

"No, open word"

"Got it. I'll close that for you. Anything else?"

[–] InevitableSwing@hexbear.net 9 points 1 week ago (1 children)

the most obscure and Byzantine menus and features

It's as if that's their goal. Why give the user the option of a single click to do what they want? That's no fun! Let's give them six inferior ways that take five times as long. Plus we should consider an extra buggy method that only works about 50% of the time.

[–] DogThatWentGorp@hexbear.net 6 points 1 week ago

My favorite is changing audio settings or plugging in new devices.

They have the device manager, then they also have the microslop minimalist block menu "your devices" thing THEN they have different sub menus attached to a whole other part of system settings for monitors and audio each THEN there's a per-app audio mixer that you can only get to by navigating through a chain of menus somehow.

And everytime you try to get to the normal, legacy device manager it sends you to the funky "these are your microphones" page that doesn't even let you detect or configure anything.

And I put up with it like a little pain pig. Like a little masochist freak. I NEEED Linux the next time I can make a PC (which will be god knows how many years at this point).

[–] Carl@hexbear.net 12 points 1 week ago

mad libs noun

[–] Le_Wokisme@hexbear.net 30 points 1 week ago (1 children)

Heavy AI users can cost up to $14,000 on a $200 subscription,

praxis?

[–] spectre@hexbear.net 7 points 1 week ago (1 children)
[–] HexReplyBot@hexbear.net 2 points 1 week ago

I found a YouTube link in your comment. Here are links to the same video on alternative frontends that protect your privacy:

[–] KnilAdlez@hexbear.net 14 points 1 week ago

I have a small llm that runs on a 10 year old GPU, it gives me a summary of the weather and my to do list twice a day, and I use it for voice commands for home assistant. If I really wanted it to summarize emails or draft responses or something, I'm sure it could handle it just fine. These giant models are basically worthless for day to day usage.

[–] Carl@hexbear.net 10 points 1 week ago* (last edited 1 week ago) (1 children)

Businesses are switching to Chinese AI providers charging 18 cents per million tokens versus $4 for Western models, reshaping the industry.

It'll be interesting to see how/if China responds to this shift. Back when crypto was booming, the strain on the grid from the increased power draw triggered mining bans in some cities. This is inherently one of the problems a modern electrical grid at subsidized prices will always have to deal with - people who use a lot of electricity all want to move their operation to where it is cheaper.

[–] jackmaoist@hexbear.net 9 points 1 week ago (1 children)

Chinese models are open weight so western data centers just run it themselves and offer them as services. I expect Chinese labs to make their models proprietary when their popularity inevitably rises.

[–] peeonyou@hexbear.net 1 points 1 week ago

I don't know that they will

[–] peeonyou@hexbear.net 6 points 1 week ago* (last edited 1 week ago)

well if i knew i could bleed microsoft dry by using copilot i might have run it a time or two