AGI Dreams Podcast – April 28, 2026

AGI Dreams – Open, Uncensored, & Local - AI News Digest · Apr 28, 2026 · 15m · transcript

The Quantization Arms Race

In this episode:

The Quantization Arms Race — The open-weight community is pushing VRAM optimization to extremes, with Qwen 3.6-27B fitting on consumer GPUs at 110K context via IQ4_XS quantization patches. Meanwhile, BitNet's ternary compression and new consumer hardware like the RTX 5090 are making large local models increasingly viable alternatives to API providers.

AI Coding Agents Level Up — Open-source coding agent Dirac topped TerminalBench-2 with hash-anchored edits and AST-native manipulation at 64% lower cost than competitors. New infrastructure…