Reddit eyes larger Gemma 4, Qwen3.6

The demand is real: people want a “flagship-but-runnable” open model that feels closer to the big closed models without blowing up local inference costs. But the current releases suggest both labs are still balancing capability against deployability, not simply scaling for bragging rights.

• –Gemma 4 is already a four-size family with a 26B MoE and 31B dense variant; Google has not officially announced a larger open MoE checkpoint yet.
• –Qwen3.6 has moved fast on both open weights and API variants, but the jump to a much larger MoE like a 122B-class model remains community speculation, not a confirmed roadmap.
• –The thread highlights a clear local-LLM preference: if active parameters stay modest, a larger MoE can be attractive; if not, the hardware cost quickly becomes impractical.
• –For developers, the interesting question is less “will they scale?” and more “will they keep the activation footprint small enough to stay self-hostable?”
• –Treat this as sentiment and roadmap-watching, not a release announcement.