signal: week of 2026-06-19
2026-06-19
https://tinksoft.com/log/signal-2026-06-19/
--------------------------

Open weights ate the week. Two big Chinese coding models, a corrected benchmark, and a couple of developer-tooling shifts worth a look.

- [Z.ai releases GLM-5.2 open weights](https://venturebeat.com/technology/z-ais-open-weights-glm-5-2-beats-gpt-5-5-on-multiple-long-horizon-coding-benchmarks-for-1-6th-the-cost) — Z.ai published the full MIT-licensed weights of GLM-5.2 on June 16 (753B params, 40B active, 1M-token context), claiming it beats GPT-5.5 on several long-horizon coding benchmarks at roughly one-sixth the cost.
- [Moonshot AI releases Kimi K2.7-Code](https://www.marktechpost.com/2026/06/12/moonshot-ai-releases-kimi-k2-7-code-a-coding-model-reporting-21-8-on-kimi-code-bench-v2-over-k2-6/) — Moonshot shipped an open-weight (modified MIT) coding model, a 1T-parameter MoE with 32B active and a 256K context, reporting +21.8% on its own Code Bench v2 over K2.6 while using about 30% fewer reasoning tokens.
- [Epoch AI ships FrontierMath v2](https://epoch.ai/benchmarks/frontiermath-tier-4-v2) — After an audit found small but critical errors in 42% of the original problems, Epoch corrected 135 and removed 12, leaving a 338-problem set on which Claude Fable 5 now holds the top spot.
- [OpenAI adds Record & Replay to Codex](https://developers.openai.com/codex/changelog) — On June 18 OpenAI added a macOS feature that turns a demonstrated workflow into a reusable Codex skill.
- [Gemini 3.5 Flash hits general availability](https://ai.google.dev/gemini-api/docs/changelog) — Google moved Gemini 3.5 Flash to GA in the Gemini API on June 16, positioning it as its default model for sustained agentic and coding work.
- [Anthropic ships a Claude Design overhaul](https://www.techrepublic.com/article/news-anthropic-claude-design-overhaul-enterprise-teams/) — A June 17 update adds design-system imports from GitHub repos or raw uploads, tighter Claude Code integration, direct canvas editing, and more export options.
