45 comments

[ 1.8 ms ] story [ 54.2 ms ] thread
open weights and

similar performance to deepseek v4, inkling at size of nemotron 3 super (!)

Amazing model at this size if true, that's quite crazy!
This is INSANE. How did they do this?
Are all AI labs in Google's weight class in crawling and ranking the Web's content? I know OpenAI has contractor subject matter experts in all topics.
Model Looks amazing!

Even more important, subjectively, is that this model will run very well on Strix Halo (e.g. Framework Desktop), DGX Spark kinds of devices. Looking forward to Unsloth dynamic mtp quants.

P.S. Looking at the HF release they already offer Q4_K_M and DFlash drafter for speculative decoding!

This is exactly the kind of model that's been needed in the middle. Realistically self-hosted, Good Enough intelligence, MoE so it's fast on limited bandwidth systems like Strix Halo and DGX Spark.

For a while there's been nothing to run on my Strix Halo that's notably better than what I can run on my dual 32GB GPU desktop (Gemma 4 or Qwen 3.6 dense models), but this seems likely to be the step up in size that actually works better than those.

On Strix Halo, have a look at the work antirez is doing with dwarfstar (antirez/ds4) for deepseek v4 flash at reasonable speeds and quality.
I love this. Is it possible to give a feel of how this stacks up to the good old Opus 4.5 in coding quality? For me that was the turning point where agentic coding in Claude Code etc became usable. Have we hit that threshold?
Whoa whoa whoa, 118b params, 8b active MOE, long context reasoning, open weights - music to my ears. Hadn't heard of this lab before but I am very excited, will definitely try this out tomorrow - this is a real sweet spot I think in terms of model size and performance.
Testing it now. At the very least, competitive with DS4-Flash indeed. On my small (and per Sol's words, _very_ semantically dense) C test codebase, it found things that only gpt-5.2 managed to find back in the day, but also made a stupidly incorrect initial observation that a memfd_create()/mmap was used for IPC (funnily enough - sol missed that as well in its review, until I pointed it out). Re: the claims vs deepseek v4 - both flash and pro are expected to get a "general availability" release very soon (i.e. well-"post-trained"), so things can change in a... well, flash, as per usual in the current environment.

Anyways, keep 'em coming.

(comment deleted)
Incredible. This is definitely the launch of the day. Just crushing Google's releases.

The pricing here is incredible. This is the first US release that's competitive with DeepSeek V4 Flash. Very excited about this.

(comment deleted)
Looks impressive, and this size fits achievable home hardware.

That said, if someone would kindly quantise this down for the 64GB paupers, that would be appreciated. (I know there’s likely degradation, but some people reported good results with a 2 bit version of Qwen 3.5 122B, and this is starting from a higher point. Would be interesting to try, at least.)

Edit: someone in the process of doing so: https://huggingface.co/vcruz305/Laguna-S-2.1-GGUF

Any estimates of the performance (prompt processing and decoding tokens/s) on consumer hardware like Macbook Pro M-series?
!! Be careful when testing the model.

A lot of people are testing it, and reporting disappointed results / benchmaxxxing claim. But do not realize that thinking has a issue with the default configuration.

Important - make sure that THINKING is enabled. By default it wasn't although I was passing the flag --default-chat-template-kwargs '{"enable_thinking": true}' in vllm recipe. The generation_config.json file that is included has by default max_new_tokens as 32k which seems to be cutting off thinking altogether so increase it. At first I was very disappointed with the output I was seeing, but once thinking is enabled, the code quality seems to be MUCH better. More real world testing to be done.

https://www.reddit.com/r/LocalLLaMA/comments/1v2pg99/laguna_...

Looks like the default chat template was updated on HF to enable this by default shortly after you posted this :-)
holy shit its accelerating fast
happy the US has some counterweights to the Chinese labs, just need about half a dozen more.
Really impressive signal that this 128B model can beat DeepSeek V4 (1.6T) on most coding benchmarks!

Also, I really like Poolside's habit to compare not only to other top models in its weight class (others don't do it, looking at you Mistral), but also to the very top open-weight models, even much bigger ones like the 2.5T Kimi-K3!

This is fantastic work, really impressive is an understatement. I really hope this sets a new DeepSeek-esque standard and starts another the death knell for companies continuing to cosplay as frontier labs (like Cohere).
Has anyone tried it on a mac yet?
Was able to run it on Apple M3 Max (128 GB)

host: Apple M3 Max, 128 GB model: Laguna-S-2.1, 118B-A8B MoE, Q4_K_M (75 GB), DFlash speculative decoding server: http://127.0.0.1:8000, llama.cpp, ctx 64K, 8-bit KV cache

  mode: max thinking
   #  tokens   tok/s  dflash
   1     600    14.4     11%
   2     600    26.1     27%
   3     600    17.8     18%
   4     600    14.0     16%
   5     600     9.3     15%
  --------------------------------
  median  14.4   mean  16.3   min   9.3   max  26.1   tok/s

  mode: no thinking
   #  tokens   tok/s  dflash
   1     190    10.0     20%
   2     109    26.7     65%
   3      95    29.6     72%
   4      93    32.8     81%
   5     382    14.0     30%
  --------------------------------
  median  26.7   mean  22.6   min  10.0   max  32.8   tok/s
You will not, directly or indirectly: copy, modify, or create derivative works of the Site or any Product;
initial impressions, great model for coding, probably swapping it out for qwen 27b for a while to long-term test, more sycophantic than any I've run locally myself
Immediate reaction is that it seems to be a bit behind Meta Muse Spark 1.1 performance at approximately the Deepseek v4 Flash price point. That's quite good given Muse Spark benchmarks a lot better than Deepseek v4 Flash (assuming benchmarks mean anything, which they don't).
This thing is great, twice as fast as DS4Flash and slightly smarter too. I swapped most of my sub-agents to this model.