ICE 2.0
Ranking: The model I found to be most capable in real-world tasks is placed at the top, and the list continues downwards using the same logic.
Text Generation • 36B • Updated • 19.2k • 9Note This model is gold standard of Qwen3.6 35B-A3B, standard practice, default start point.
gbuzhf/Nex-N2.5-mini-OrcaRouter-Sangreal-ICE
Text Generation • 35B • Updated • 1.76k • 4Note Pretty good, useful and smart for coding, also good for "daily" use.
gbuzhf/MiniCPM-V-4.7-35B-A3B-Uncensored-Sangreal-ICE
Text Generation • 35B • Updated • 4.58k • 9Note Fast on jobs, but never eager to dive deeper. Useful for daily/mobile/light tasks (even with complexities); Not useful for specialized coding/reversing/sophisticated engineering, etc. tasks.
gbuzhf/Apex-35B-V1.0-Uncensored-Sangreal-ICE
Image-Text-to-Text • 36B • Updated • 7.17k • 3Note Benchmarks reaching to the stars! But only for BF16; Ultra-sensitive to quantization. Good on "repeat until solve" type coding tasks, obsessive and heavy-lifter. Not useful for day-to-day tasks; Can try to write an engine to calculate 2+2, and two more diagnostics scripts for the engine's maintenance.
-
gbuzhf/Sage-A1.5-Sangreal-ICE
Image-Text-to-Text • 36B • Updated • 14.8k -
gbuzhf/Apex-35B-V1.0-Sangreal-ICE
Image-Text-to-Text • 36B • Updated • 3.05k -
gbuzhf/SNEX-V1
Image-Text-to-Text • 36B • Updated • 5.67k • 2