Many Downvoted me for saying this a while ago. Qwen 3.7 released with no Open models.
Posted by MLExpert000@reddit | LocalLLaMA | View on Reddit | 38 comments
You may not like what I say, or it may hurt your feelings. But the truth is truth.. end of the day it’s about profits. But you can feel free to downward me because you have the freedom.
Finanzamt_Endgegner@reddit
Bro they gonna release oss models, their engineer even said 27b will probably be ready soon
MLExpert000@reddit (OP)
Where?
Finanzamt_Endgegner@reddit
Twitter I can't remember his name but he replied to some guy asking below the official qwen account if mz memory serves me correctly.
sleepingsysadmin@reddit
Qwen has always had closed mega models that only run via datacenters. That's not new or changing.
Implying that they've given up on open source is absurd.
MLExpert000@reddit (OP)
Same argument was made before . What I’ve been saying is it’s inevitable they become closed . They might release some oss here and there for the sake of it so that ppl won’t hate them
TheOriginalG2@reddit
they have no reason to release once they have market share.
TheOriginalG2@reddit
How about now? is it still absurd? I don't see a drop and it's been.... A WHILE. Not even a peep.
rpkarma@reddit
We’ll see though. They didn’t release many 3.6 models, and I won’t be surprised to see no 3.7 models at all. I hope I’m wrong and you’re right!
jld1532@reddit
Releasing open weights is a core strategy in collapsing American for profit API. There will be more open models.
MLExpert000@reddit (OP)
Hope so. Many more and smaller
ghac101@reddit
Don't think so. There always have been open source and close source models. Think they just partner for the closed ones and continue open sourcing. No one realistically will self host a 1 trillion parameters model. Therefore, does not make sense to open source.
-dysangel-@reddit
This is the worst sub to be making such bold claims lol. If you said "not many will realistically self host a 1 trillion parameters model" you'd be more accurate.
ghac101@reddit
Basically meant very view by saying no one 😂 there's always a unicorn out there.
-dysangel-@reddit
I've run Kimi locally before 😃
suicidaleggroll@reddit
Qwen always releases the Pro first, then releases the weights for the smaller versions later. It's happened every release. Do you have any evidence they won't release the weights for smaller versions in the next few weeks?
MKU64@reddit
Never. They always released Max AFTER the open/source models and Plus was always open-source. They started doing it after Qwen 3.6 which was when they started to be more closed-source.
MLExpert000@reddit (OP)
Exactly. This is a recent patten after they shuffled their team internally.
-dysangel-@reddit
Yep, pretty interesting that people don't notice the details. I'd be very happy to be proven wrong, but things are not looking good so far.
Finanzamt_Endgegner@reddit
"Plus was always open-source." it wasnt since 3.5
"They always released Max AFTER the open/source models" They released the last one for 3.6 before 27b.
MKU64@reddit
Qwen 3.5 Plus was open-source. They even tweeted that it was just 397B with their 1M context technique which they also open-sourced.
“They released the last one for 3.6 before 27b”: That’s exactly what I said?? Everything started with 3.6 which was when the Qwen Team was swapped internally.
JsThiago5@reddit
From what I saw since I entered the hobby, it is the other way around.
EmPips@reddit
I don't want to get in the middle of it but isn't this wrong?
Qwen2.5 Max released 3 months after Qwen2.5 weights
Qwen3 Max released 5 months after Qwen3 weights (2 months after the 2507 updates)
Qwen3.5 is a bit fuzzier as the release was paired with a closed source version built for higher contexts.
Qwen3.6 Max Preview came out after 35B but before 27B.
Again I don't take any of these signals seriously but I don't think there's any precedent that Qwen releases closed models to ramp up for an Open Weight release.
LetsGoBrandon4256@reddit
You do know that Max and Plus has always been proprietary, right?
TKGaming_11@reddit
Qwen3.5 plus was open weight as Qwen3.5-397B
ortegaalfredo@reddit
Its not the same model. Qwen 3.5 had longer context.
TKGaming_11@reddit
It’s the same model with yarn context extension, the details are listed on the model card
LetsGoBrandon4256@reddit
It's fucking joever.
SnooPaintings8639@reddit
"Many Downvoted me..." - and many more will!
MLExpert000@reddit (OP)
Please do. 🙏
pmttyji@reddit
I'm sure we'll see open-weights of small/medium models in June/July & big/large models in July/August.
Klutzy-Snow8016@reddit
Kimi and DeepSeek are also available on Fireworks, so I'm not sure what you're trying to say with that screenshot.
MLExpert000@reddit (OP)
The devil is in the details
vergogn@reddit
EbbNorth7735@reddit
Wtf... obviously it's about profits and they don't need to release open source models. That said it's highly unlikely they won't release open source software.
complexminded@reddit
The reality is that more people should act like these are the last open models we are getting, and utilize them to the best of their ability, instead of constantly looking for the next release.
I operate as if qwen/minimax/etc is the last model I'm getting. Because it could be, you never know.
0-0x0@reddit
That's the logical approach, current open weight models are the last until proven otherwise.
Wrong_Mushroom_7350@reddit
People will downvote this because it’s hard to admit.
So I always reframe it like this. You made a product, that’s open source and someone came to you and offered you “400 million dollars, that you can deposit in your bank.”
Would you really sit here and turn down 400 million dollars?
Some people will write and grand stand that under no circumstances, they would take the money.
03captain23@reddit
Wow