- Author
- Nathan Lambert (Interconnects)
- Published
- 22 June 2026
- Kind
- Newsletter opinion post with brief hands-on use
- Original
- www.interconnects.ai/p/glm-52-is-the-step-change-for-open
- Subject
- Harnesses and agent loops
- Themes
- Who holds the power, Given away
What it is
Nathan Lambert writes the Interconnects newsletter on AI. This post gives the author's reaction to the release of GLM-5.2 from Z.ai. It places the release in the open-closed gap and the regulation debate.
What it says
- Z.ai released GLM-5.2 to plan members on 13 June and published weights on 16 June.
- Lambert calls GLM-5.2 the first open-weight model that feels right in coding harnesses.
- Arena's agent leaderboard had GLM-5.2 as the only open model near the latest closed models.
- The release gap from Claude Opus 4.5 to GLM-5.2 is 204 days, about 6.8 months.
- Lambert expects pricing pressure on closed labs and a boost for open-model inference sellers.
- Lambert warns the US government might deem an open Chinese model unsafe, among other scenarios.
- Lambert used it in Claude Code through Fireworks, where images could break a session.
What evidence it gives
The post cites community leaderboards and praise from named people. It reports brief personal use in Claude Code through the Fireworks API. It also works out a 204 day gap between two release dates.
Interests and limits
This is an opinion post, with no benchmark numbers or tests of its own. Personal use was brief and the author says tinkering remains. The author writes about open models and promotes related paid subscriptions and a new paper. The voiceover was not heard.