30B Video MoE Quantizations Tested: Most Users Should Wait

EntireBig7258 · reddit · 2026-08-03

The author compares the mainstream community quantization builds for the LingBot-Video 30B-A3B MoE model:

Conclusion: The tooling is still stabilizing, and 30B quants have documented quality costs. Most users should stick to running the 1.3B model at full precision for now.

Original post →

More from Infra

Infra channel →