AI Agents Just Changed Forever: GLM 5.2 The Best Open-Source Model Ever?
TL;DR
GLM 5.2 is the open model Riley thinks is finally real: He says Z.ai's model is far cheaper than GPT 5.5, felt comparable in real use to Claude Opus 4.8 and GPT 5.5, and handled coding, research, Notion, and Slack tasks without the usual benchmark hype letdown.
You can add GLM 5.2 to Cursor in 3 to 5 minutes with OpenRouter: Riley shows a simple setup by overriding the OpenAI base URL, pasting an OpenRouter key, and adding the custom model
z-ai/glm-5.2so it works directly inside Cursor agent sessions.Codex can now turn a screen recording into a reusable skill: Using the new record and replay plugin, Riley records a Typefully draft workflow, and Codex converts that recording into a callable skill that later tries to upload a video draft automatically.
SpaceX buying Cursor matters because of resources, not headlines: Riley's take is practical. The reported $60 billion deal could give Cursor more compute, more capital, stronger distribution through Elon Musk's network, and a better shot at competing with Codex and Claude.
People are in a real 'Fable depression' after Anthropic pulled the model: Riley says users who briefly had Fable access now find Opus 4.8 and GPT 5.5 less ambitious by comparison, and he frames the strongest benchmark as whether a model keeps attacking hard problems instead of backing off.
Claude's newer features point toward agent-native mini apps: He highlights web-only design mode upgrades like direct canvas editing and deployment to Vercel, plus Claude Code artifacts that generate shareable interactive pages for teammates.
The Breakdown
GLM 5.2 might be the first open-source model in a while that actually "passes the vibe check". Riley Brown says it holds up surprisingly well against GPT 5.5 and Claude Opus 4.8, then walks through how to wire it into Cursor in minutes while also hitting Codex's new screen-record-to-skill feature, SpaceX's Cursor acquisition, and Anthropic's strange post-Fable slump.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.

Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.

Playbook
The Art of Tasteful Prompting
Learn how tasteful prompting helps you move beyond generic AI output by shaping context, style, and judgment from the start.