Portable Agent Plugins searchable across local, workspace, and remote catalogs, MCP 2026-07-28 protocol with paginated discovery and multi-round requests, and Amazon Bedrock caching all land in one release; secret redaction from displayed commands also added.
Why it matters: MCP support means Codex can now plug into the broader tool ecosystem operators build on — agent workflows built for one platform become more portable across hosts.
A new classifier fixes false positives that previously dropped Fable 5 to a weaker model on biology queries, restoring full capability for lab result interpretation, symptom questions, and healthcare practitioner tasks.
Why it matters: Any workflow touching health or life-science data can now rely on Fable 5 without the surprise model-downgrade tax — practitioners in regulated industries get more consistent output quality.
xAI's new Quality Mode on Grok adds region-level editing (magic wand, segmentation, background removal), multi-reference inputs up to 5 sources, and Smart Resize; ranks second behind GPT-Image-2 on both text-to-image and image editing benchmarks, with API access coming soon.
Why it matters: Grok's image model is now production-ready for brand-consistent creative work — multi-reference generation closes the gap between AI ideation and usable ad creative.