GLM 5.2 vs. Opus (techstackups.com)

Article summary

The article compares the performance of GLM-5.2 and Claude Opus 4.8 on a task to build a 3D platformer game from scratch in raw WebGL. GLM-5.2 took longer and produced a rougher game, but at a fraction of the cost. Opus finished faster and produced a cleaner game, but at a higher cost. The article also discusses the benchmarks and online reaction to GLM-5.2.

Main themes

• LLM comparison
• Game development
• Model performance
• Cost and pricing
• Benchmarking
• Agentic tasks

What commenters say

• A single one-shot prompt is not a reliable benchmark for comparing LLM performance.
• The test was not a formal benchmark, but rather a vibe test to compare the models' capabilities on a long-running task.
• The cost comparison between GLM-5.2 and Opus should include subscription prices, not just pay-as-you-go pricing.
• Some commenters argue that GLM-5.2 is not as efficient or steering-friendly as Opus, but produces good output.
• Others suggest that the subsidy status of the models is irrelevant to consumers, who only care about the price they pay.
• There is disagreement about whether the API prices of major labs are indicative of their profitability or subsidization of subscriptions.
• Some users are considering switching to GLM-5.2 or combining it with other models to reduce costs.
• The lack of vision support and slower speed of GLM-5.2 are concerns for some users, despite its lower cost.