If you're running Laguna S 2.1 and it feels "stupid" or isn't reasoning properly

<!-- SC_OFF --><div class="md"><blockquote> <p>Wait, I need to be careful about the &quot;no_think&quot; tag in the system prompt. The system prompt says <code>/no_think</code>.

<!-- SC_OFF --><div class="md"><p>I am currently running the new Laguna model at Q2_K_XL on dual 3090s for reference, with a Q8 context of 200,000. </p> <p>Using Pi, I am noticing

<!-- SC_OFF --><div class="md"><p>I dont have good experience running q4_k_m, the difference to q6 is &quot;a few errors an hour&quot; to &quot; a few errors every couple of

<!-- SC_OFF --><div class="md"><p>I could fit the Q3 model not sure if it's worth over 27B ? </p> </div><!-- SC_ON --> &#32; submitted by &#32; <a

<!-- SC_OFF --><div class="md"><p>Currently using glm 4.7 flash, it’s very meh </p> <p>Heard omnicoder or Crow 9b are good, are they any better?</p> <p>Or Qwen3.5 27b?</p>

<table> <tr><td> <a href="https://www.reddit.com/r/LocalLLaMA/comments/1v3kvgz/lagunas21_failed_basic_intelligence_litmus_test/"> <img alt="Laguna-S-2.1 Failed Basic Intelligence

<!-- SC_OFF --><div class="md"><p>basically the title. recently Ive tried Laguna on openrouter (since its free). And compared to the nemotron ultra 550b I was surprised how Laguna

<!-- SC_OFF --><div class="md"><p>That foodtruck bench post showing deepseek v4 matching gpt-5.2 at 17x cheaper got me thinking. if frontier cloud models are that overpriced for