Flash Next is now first-class in llama.cpp. The catch if you have an M3 Ultra: you do not get the M3 Max Metal flash-attention tables.
Friday wrap of AI Engineering Digest Issue 6:
youtu.be
AI Engineering Digest — Issue 6 (Aug 28, 2026)
Friday wrap of this week's written AI Engineering Digest.
Full issue: https://ai-eng-digest.astgl.ai
0:00 Intro
0:10 llama.cpp
0:30 Metal flash-attention
0:56 Like and subscribe
1:07 Bedrock AgentCore
1:21 Nemotron 3.5
1:36 mlx-lm
1:53 Flash Next on Mac
2:16 MiniMax-H3
2:33 GLM-5.3-Flash
2:50 Open
