Wednesday, September 9, 2026

vLLM 0.29.0 goes GA.

After Tuesday Issue 12 close, vLLM 0.29.0 landed on GitHub and PyPI Wednesday morning — Model Runner V2 default for all models, Hy4 / Qwen3.8-Flash-Next / Kimi K3 NVFP4 support, and breaking upgrade notes (FLAG project benches). AWS What’s New Sep 8 shipped AgentCore Memory IngestData — long-term ingest without short-term events — plus Agent Gateway VPC connectivity templates and Feature Store UpdateRecord. llama.cpp b10863 Metal iq3_xxs idle-thread fix is the local lead. No new frontier LLM API id; OpenAI Images 2.5 is Other (image stack). MLX quiet (still mlx-vlm 0.7.0 from Issue 12).

Window after Issue 12 close (~2026-09-08 morning ET) through 2026-09-09 morning. Issue 12 already covered mlx-vlm 0.7.0. Primary sources only. Vendor evals flagged.