Laguna S 2.1

“the Laguna architecture didn't lend itself to the same kind of dissection as Qwen3.8”

“I think this is probably also stupider than laguna s 2.1 which can also be very cheap to serve”

“their latest updates means it doesn't quite fit a GB10 128GB anymore at full context which is a shame”

“I wasn't blown away by its capabilities, it's also totally acceptable for many purposes”

“I found a couple of different 4-bit quantizations of Laguna S 2.1 that run pretty well with pretty big context”

“they borked the quants released with the main model and had to update the next day”

“Laguna and Muse Glimmer failed hard on it, though they're useful for other tasks.”

“Laguna and Muse Glimmer failed hard on it, though they're useful for other tasks”

“Another impressive recent model is Laguna S 1.2 .. about as good as DeepseekV4 in general but with more refined design sensibilties”

“For me Laguna frequently slightly corrupted text then it would be unable to notice the difference”

“Works amazingly for coding, pretty bad for pretty much anything else.”

“Initially there were serious bugs, also about the templates”

“seems to be on par (or better) than DeepSeek v4, is only 110B, and it's faster”

“pretty bad as an agent - not proactive, would frequently stubbornly argue things that weren't true”

“There was an obvious problem in the original release, they re issued it after like a week with the reasoning looping supposedly fixed.”

“pretty good (not frontier, but competitive with the tier 2 models)”

“laguna s 2.1 I think is the best ~100B moe model right now”

“things like Laguna S2.1 NVFP4 gets really close of almost being there, when it comes to coding specifically”

“I am SO impressed by this model! It's cheap as deepseek v4 flash, better than minimax m3 I would say”

“the outputs so far aren't impressive, and it gets stuck and perseverates”

“very memory hungry and PP speed drops off a cliff around 200k context”

“it positively flies on an m5 max - the 60tok/s decode / ~500tok/s prefill is real.”

“Laguna provided quants have some issues and they are reworking/updating them.”

“very comfortable for chatting, but still not usable for interactive agentic coding”

“Looks to be working fine, but too early to tell. Definitely likes to "think".”

“This is really a sweet spot imo on this machine between maximizing ram use and still having decent speed due to the expert size.”

“Seems quite literally they have bugs on the hardware I'm trying to run this with”

“impressed enough so far I'm going to see if it can replace my qwen3.6 27B usage”

“have only tested a few prompts, but it failed my favorite non-coding question”

“it seems to be a bit behind Meta Muse Spark 1.1 performance at approximately the Deepseek v4 Flash price point”

“Usually new model releases are plagued with issues at release though”

“Really impressive signal that this 128B model can beat DeepSeek V4 (1.6T) on most coding benchmarks!”

“once thinking is enabled, the code quality seems to be MUCH better”

“Looks impressive, and this size fits achievable home hardware.”

“this model is not a joke! Exciting, we already got a usable PR of work out of it.”

“Laguna S 2.1 (released today) just destroys Flash-Lite underly and completely.”

“It is doing good but at one point i had to stop it and say 'you're overthinking this' haha”

“Having played with it for like 3 hours now....I'm probably moving from CC to this”

“this is a real sweet spot I think in terms of model size and performance”

“MoE so it's fast on limited bandwidth systems like Strix Halo and DGX Spark”