Open Models - May 2026

Posted by pmttyji@reddit | LocalLLaMA | View on Reddit | 29 comments

[After overwhelming April](https://www.reddit.com/r/LocalLLaMA/comments/1t06y43/open_models_april_2026_one_of_the_best_months_of/), May seems underwhelming even though we got Ring, Command, StepFun, LFM models. Hoping for great June(We're getting MiniMax-M3 in 10 days). ^(PS : Took me 15-20 mins to gather these models & generate this graph. BTW this graph is not a benchmark)

Reply to Post

29 Comments

[-]

YutaBye@reddit

Thanks for your share.👍

[-]

srigi@reddit

Do we still believe to be gifted with Qwen3.7-27B?

[-]

pseudonerv@reddit

No. Nothing is free. They have gone where meta went.

[-]

--Spaci--@reddit

qwen doesn't release a model for a month and mfs say shit like this 😭💔 If I was qwen I wouldn't want to release models to you savages either

[-]

BannedGoNext@reddit

They literally said they would release it on X, bunch of fucking doomers. They will probably say when they change and are done releasing OS.

[-]

pseudonerv@reddit

You want to dream? Me too. But they clearly didn’t even train any small models for 3.7. And rumor has it that the released two 3.6 models were some left overs before their restructure. The restructure is simply oriented towards monetizing. So basically where llamas went.

[-]

--Spaci--@reddit

we will see

[-]

pmttyji@reddit (OP)

Soon or later for sure.

[-]

No-Refrigerator-1672@reddit

Do we believe that Qwen3.7 weight will be released **before** the model gets outdated?

[-]

patricious@reddit

I believe they will bless us with it soon, yes.

[-]

kiwibonga@reddit

As soon as 3.6 4B and 9B come out

[-]

Zestyclose_Potato794@reddit

Still wondering what the lfm model can do if used as an util model. Did someone try this ?

[-]

mxforest@reddit

You forgot to add Nemotron Ultra?

[-]

Dany0@reddit

Doesn't come out until June 4th. Also it was announced on June 1st

[-]

Real_Ebb_7417@reddit

Oh man, thank you, I need more summaries like this every month 😃

[-]

pmttyji@reddit (OP)

Sure, will do.

[-]

Technical-Earth-3254@reddit

Can we talk about how 50 million isn't 0.5 billion? lmao wtf

[-]

Nnyan@reddit

Always that damn missing 0.

[-]

pmttyji@reddit (OP)

My bad. Can't replace the image.

[-]

Tall-Ad-7742@reddit

nah fr this can't be true ring is so bad it is never better then a 0.5B model... typical trust me bro benchmark /s

[-]

pmttyji@reddit (OP)

Next time onwards, I'll put text on graph itself. It's not a benchmark graph. Just models & its parameters.

[-]

Tall-Ad-7742@reddit

relax /s stands for sarcasm... it was just meant as a joke

[-]

pmttyji@reddit (OP)

Didn't notice. Even on April month thread, few people mentioned this so replied.

[-]

Achso998@reddit

Did you even looked at the graph? It's not a benchmark...

[-]

Tall-Ad-7742@reddit

relax /s stands for sarcasm... it was just meant as a joke

[-]

Achso998@reddit

Oh my bad sry. Didn't realize that

[-]

buttplugs4life4me@reddit

Reddit says Theres 10 comments here none of which are visible, which means either they blocked me, I blocked them, or they're shadowbanned. Weird. I'm looking forward to Zaya and Intern-A honestly. Both seem the most interesting. I may see what StepFun Quants down to if I can I use it. Never heard of Ovis so thanks for that

[-]

Imscomobob@reddit

I think sometimes Reddit just takes forever for comments to actually show up…

[-]

anykeyh@reddit

I'm excited for zaya 72 once the RL is done. It promises qwen 27B performance in a MoE 4B parameter. Excellent for strix and Mac studio.