Anthropic's Claude remote uses GLM-4.7

Posted by bobbiesbottleservice@reddit | LocalLLaMA | View on Reddit | 45 comments

I just noticed this after a bug wasn't getting fixed. If you start a Claude code remote environment the default model (hidden on mobile) is glm 4.7 I assumed anthropic only used their own models for everything so it was interesting to me that they are actually serving open weights models (which I assume they have accused of distilling from their own proprietary models)

[-]

Ell2509@reddit

Are you sure you didn't point CC at it, somehow?

If so, either you just blow the lid off some global scale plagiarism, or you found a clue about current or future commercial collaborations between the 2 labs.

[-]

Likeatr3b@reddit

Can you do this yet? I’ve predicted that local AI will eat their lunch. But haven’t seen them fold and allow local models to use their products.

[-]

ThoreaulyLost@reddit

I’ve predicted that local AI will eat their lunch.

That's a very, very, very optimistic prediction. While I personally would love to see that, there are too many political and financial forces in play that will prevent that from happening at scale. Just look at the chip and drive shortages.

The reason corporate AI will win is because old money us entrenched in it. They don't like not getting their investment back. Right now we're in the "let's pretend there's competition" (think 1980s computers, or early car manufactures) when it's in reality a pretty incestuous little network of "leaders". Given 20 years it will be a consolidation of monopolistic partnerships, who probably divide the field like kings after a battle (lower AIs in cars or vacuums, medium AIs in civilian products, top AIs in government surveillance, traffic control, aiports, etc).

My long term bet is on whomever does the next AI merger, which is why, if OP is right, I'm off to buy some stocks.

[-]

Ell2509@reddit

Sadly, you are right.

Anyone who is unsure or hasnt heard of it, Google "techno feudalism"

[-]

yinepu6@reddit

or google "google techno feudalism"

[-]

imonlysmarterthanyou@reddit

You have been able to for a long while.

[-]

bobbiesbottleservice@reddit (OP)

I was using clother with Claude code in another terminal (not on a remote session), but I never used glm-4.7 and I had other local models I was using too, but none of them showed up.

[-]

jatjatjat@reddit

That lower case g not lining up with the other model names is sus.

[-]

AnotherSoftEng@reddit

Don’t be ridiculous, even Anthropic’s senpai model does this!

[-]

john0201@reddit

It’s not a fixed width font. No idea if this is real or not but the font looks real.

[-]

_supert_@reddit

Why has this got 68 upvotes? With no evidence?

[-]

mp3m4k3r@reddit

And what does it have to do with Local other than you could potentially run GLM locally?

[-]

my_name_isnt_clever@reddit

"Closed cloud provider using open weights model" is absolutely relevant to this sub...if it were true.

[-]

mp3m4k3r@reddit

Good call, on both parts!

[-]

Enough_Big4191@reddit

yeah wouldn’t read too much into that yet, these setups often mix models behind the scenes depending on task, latency, or cost. also worth double checking it’s not just a label or fallback behavior. i’ve seen cases where the exposed name doesn’t fully match what’s actually running.

[-]