daily ritual at this point…

[-]

spaceman_@reddit

In about 10 hours another Qwen is getting release, if the previous releases are anything to go by (they have released the other 3.6 models on Wednesday 5-7AM UTC).

50% chance it's our day tomorrow OP!

[-]

Makers7886@reddit

It was 7 days between the last two and we are on day 6. I'd guess within 24 hours.

[-]

spaceman_@reddit

Both releases were Wednesday morning UTC. We're likely to see the release of either 122B or 9B in the next 12 hours.

[-]

NegotiationNo1504@reddit

What about small tiny ones. 9-7b 2-4b 🤤

[-]

eesnimi@reddit

I wish it had some cyber-magic to fit it inside 64gb system ram with q4_k_m instead of iq4_xs

[-]

RazzmatazzReal4129@reddit

It's up. https://huggingface.co/Qwen/Qwen3.6-122B-A10B

[-]

HomsarWasRight@reddit

Oh man, this is like smarter than frontier models!

[-]

maglat@reddit

Already downloaded, works very great!

[-]

some_user_2021@reddit

It seems as good as Qwen3.6 397B-A17B !

[-]

synth_mania@reddit

Fucker

I opened that in class, volume up xd

[-]

spaceman_@reddit

You only have yourself to blame for that one.

[-]

DocMadCow@reddit

You missed the news everyone has a GPU now with Intel driver change for allocating more memory to VRAM. Intel Celeron with 48GB of VRAM and large models GO GO GO.

[-]

jzn21@reddit

I am waiting on Qwen 3.6 397b. Can't wait, have high expectations.

[-]

Aggravating_Pinch@reddit

let me know too. saves me the trouble of doing this

[-]

-dysangel-@reddit

What if we're all refreshing localllama and we've not got anyone checking hf?

[-]

waitmarks@reddit

You think a subreddit about running local models doesnt have a few agents checking constantly ready to automatically make a post as soon as something gets uploaded?

[-]

-dysangel-@reddit

I see my attempt at absurdist humour was not as obvious as it could have been

[-]

Cool-Chemical-5629@reddit

Jacek is most likely checking HF on the cron job and auto posting whenever a new model drops.

[-]

SnooPaintings8639@reddit

From my experience here, I can tell that within 1 minute after hf show this model, there will be at least 3 separate threads here with dozens of upvotes.

One minute.

[-]

Blues520@reddit

Do you think it will be better than 27b in coding since it will be moe vs dense?

[-]

dkeiz@reddit

it will be faster, like 3 times faster.

[-]

given 3.5 27b and 3.5 122b are in the same ballpark, I would bet it will probably be about the same, maybe slightly better if they trained for longer, which is all I need. I care more about the performance uplift on Macs or similar large-memory systems you get from using the MoE instead of dense.

[-]

Cleric07@reddit

Me asf

[-]

localizeatp@reddit

9B*