3 ms·
If your framework desktop is the 128G Strix Halo, I recommend giving Qwen 3.5 122B-A10B a shot. This Q5_K_M quant should be near lossless and fit with full 256
by tarruda 4mo ago
If your framework desktop is the 128G Strix Halo, I recommend giving Qwen 3.5 122B-A10B a shot.
This Q5_K_M quant should be near lossless and fit with full 256K context in about 100GB of RAM: https://huggingface.co/AesSedai/Qwen3.5-122B-A10B-GGUF https://huggingface.co/AesSedai/Qwen3.5-122B-A10B-GGUF
- Catloafdev 4mo ago3.6 scores better on coding across the board. Edit: specifically Qwen 3.6 27B beats that on coding and agentic workflows.
- tarruda 4mo agoVibe thinker also beats Opus 4.5
- bakies 4mo agoI'll keep this in mind.