Qwen3.6-27B Tests Strix Halo 128GB Limits

Strong signal that this model is interesting specifically because it sits in the local-self-hosting sweet spot, but the real question is whether long-context usage is practical on consumer hardware.

• –The model’s appeal is density: a 27B dense checkpoint is small enough to be locally relevant, but still capable enough to attract serious workloads.
• –The hard part is not just loading weights; 256K context pushes KV cache and memory bandwidth, which is where Strix Halo users will care most.
• –This is less about raw benchmark bragging and more about sustained interactive performance under long prompts, tool use, and iterative coding.
• –The discussion suggests buyers want evidence from actual owners before committing time or money to a platform-specific setup.
• –Likely outcome: workable for shorter or moderate contexts, but 256K on 128GB will depend heavily on quantization, runtime, and how much headroom the rest of the system leaves.