3 ms·
Artificial Analysis ranks qwen3.6-27b higher than qwen3.5-122b-a10b on both intelligence and coding. Does that run counter to your experience?
by apitman 2mo ago
Artificial Analysis ranks qwen3.6-27b higher than qwen3.5-122b-a10b on both intelligence and coding. Does that run counter to your experience?
- nekitamo 2mo agoThis tracks, in my experience the 27B is better at coding and instruction following. I'm shocked at how much of a difference the dense models vs MoE makes. But it's a moot point, because for local inference on consumer hardware, the MoE is so much faster.