Skip to content
Source · Daily Brief

AI Daily Brief — 18 June 2026

The GLM 5.2 hype crossed into the “marvel” zone — Jeremy Howard called it “at least as good as Opus 4.8 and GPT 5.5” and Ollama doubled its cloud GPU capacity to handle demand. Simon Willison shipped Datasette Apps, GitHub Copilot signalled the Opus 4.6 fast tier is going away, and Zhipu announced AA-Briefcase as a new agentic benchmark.

Top stories

  • Jeremy Howard: GLM 5.2 is at least as good as Opus 4.8 and GPT 5.5. The fast.ai founder’s all-day endorsement triggered a fresh wave of cloud-capacity stress for open-weights providers. via @jeremyphoward
  • Datasette Apps lets you host custom HTML applications inside Datasette. Simon Willison’s first announcement of a long-running app surface for the SQL-publishing platform. via Simon Willison
  • GitHub Copilot announces deprecation of Opus 4.6 fast. A signal that the Anthropic-model menu inside Copilot is contracting around newer tiers. via GitHub
  • Zhipu announces AA-Briefcase, an agentic-knowledge benchmark. Framed as “the benchmark for the next era of agentic knowledge work” by Artificial Analysis. via Z.ai
  • Ollama doubles cloud GPU capacity for GLM 5.2. Operator-side confirmation that GLM 5.2 has shifted real production traffic. via Ollama

Who shipped

AWS added a SageMaker detailed-metrics + Insights dashboard for generative-AI inference monitoring on CloudWatch. Kling AI rolled out Kling 3.0 Turbo on more partner surfaces (SocialSight). OpenAI’s Brockman highlighted o3 helping families with rare genetic disease — a year-old model on a public-good use case.

Open-source pulse

GLM 5.2 momentum kept compounding — Hugging Face, Ollama, and individual community voices all pushed the same line: open weights now compete head-on with proprietary frontier on text-only tasks. Howard noted the one big gap is vision — GLM 5.2 is blind, and fixing that would close the remaining moat.

Quiet corners

OpenAI shipped no models. Mistral, Meta (Llama), DeepSeek were absent from the cycle. Google DeepMind appeared only in tooling adjacencies.

By the numbers

  • 82 stories tracked across 25+ sources
  • Most-mentioned model: GLM 5.2
  • Most-mentioned lab: Z.ai / Zhipu
  • Notable absences: Mistral, Meta, DeepSeek, OpenAI releases

Compiled by AI Feed’s editor from all 82 headlines published on 18 June 2026.