TL;DR
The creator benchmarks GLM 5.2 (a new open-weight Chinese AI model from Zhipu AI) against Claude Opus 4.8 by having both build three browser games from scratch using Ollama and Claude Code. GLM 5.2 fails on the first game due to an API limitation, ties on the second, and loses on the third, making Opus 4.8 the clear practical winner despite GLM 5.2's impressive benchmark numbers.
Key points
1
GLM 5.2 was released on 16 June 2026 as an open-weight model that nearly matches Claude Opus on several benchmarks and took first place for design, beating Anthropic's own models.
2
GLM 5.2 was run locally via Ollama (single-line install) and tested head-to-head against Claude Opus 4.8 running in Claude Code across three identical game-building prompts.
3
GLM 5.2 failed the neon arena shooter entirely due to an 'image input not supported' API error, while Opus 4.8 produced a fully playable game in one shot.
4
GLM 5.2 edged out Opus 4.8 on the block breaker game (power-ups, cleaner UI), but Opus 4.8 won the synthwave endless-flier game with noticeably better 3D visuals.
5
The creator promotes Tank, his own AI orchestration framework that unifies Claude Code, Codex, Grok, and Antigravity (Google's new CLI) into one interface, available to community members.
Actionable insights
→
To run GLM 5.2 locally, install Ollama with its one-line installer and run: ollama run glm4:cloud — no API key or cloud subscription needed.
→
GLM 5.2 does not support image input in its current API, which will silently break tasks that involve vision or image context — check your use case before switching.
→
Benchmark numbers alone are misleading: GLM 5.2 scores near Claude Opus on paper but underperformed significantly in practical, creative coding tasks in this head-to-head test.
→
For parallel AI coding workflows, spinning up multiple Claude Code sessions in separate directories lets you run and compare models side-by-side in real time.
Notable quotes
“GLM 5.2 dropped on the 16th of June. Does it live up to the hype? In my opinion, no.”
“We don't write prompts anymore. We write a loop and the agent executes overnight or over days and gives us the result we want.”
“Claude Opus 4.8 won on two of the games we designed in this video and GLM 5.2 only just edged it on one game.”
Worth watching?
⏭️
Worth watching the full video?
The key findings are all here — skip the video unless you want to see the actual gameplay footage side-by-side, which is the main thing the summary cannot fully replicate.
Saved you some time? The creator still deserves a like.
Want this for your own channels?
Add the channels you follow. Every new video summarised and in your inbox the moment it drops. From £4/month.
Try it free