Content tagged with #local llm

2 items

MoE vs Dense on a budget GPU: How Qwen3.6-35B lost to a 9B Model from the same family

Apr 17, 2026

I benchmarked Qwen3.6-35B-A3B with CPU offloading against Qwopus-9B on 26 coding tasks. The MoE model has 3 billion active parameters — fewer than the dense 9B. It still got crushed. Here is why.

#AI#Local LLM#Developer Experience

Why I Started Using Local LLMs

Apr 07, 2026

When Anthropic's Claude Code started consuming tokens 4x faster than expected due to a bug, and peak-hour throttling made paid subscriptions unreliable — I realized it was time to stop depending on a single vendor.

#AI#Developer Experience#Local LLM