XHToken/Spark-X2.5-4B

Text Generation • 4B • Updated Sep 15 • 29.2k • 1.44k

Note Fast, viable bare-minimum for resource-constrained systems. Good LORD does it overshadow Qwen on the excessive reasoning output. Takes forever because it spends forever reasoning, but can seemingly do real work, eventually.

ProCreations/grug-v1.1-qwen-3.8-27b

Text Generation • 27B • Updated Aug 18 • 103 • 20

Note Quite prone to looping, wasn't able to get much useful work done

prism-ml/Ternary-Bonsai-2-27B-gguf

Text Generation • 27B • Updated 2 days ago • 1.71M • 2.95k

Note Pretty good. With spec decode can be faster than other 27B Qwens. Weaker at valid solutions but so far not terribly so; perhaps a reasonable compromise for memory-constrained environments, may not be worth it on others.

ukisai/Swift-Qwen3.8-27B-GGUF

Image-Text-to-Text • 27B • Updated Oct 1 • 491k • 476

Note Overall solid model, pretty good, doesn't overthink, not fast on generation and lacks mtp

agentionai/Signal-3.8-Flash-Next-GGUF

Image-Text-to-Text • 177B • Updated 8 days ago • 50.4k • 41

Note Pretty good. Much slower than the ISTA-DASLab GSQ-RCO builds but feels like it has a bit more clarity