GLM-5.2: A Real Competitor to Claude or Just Loud Marketing?
Chinese company Z.ai has released its new flagship model GLM-5.2, and the community has already dubbed it the "killer" of Anthropic's Claude. I have carefully studied the specifications, benchmarks, and reviews to understand how justified this title is.
What is GLM-5.2 and why is it interesting?
GLM-5.2 is an open-source model with an MIT license, which is already a significant advantage. Its key feature is a context window of 1 million tokens that does not degrade during ultra-long sessions. This allows the model to keep huge amounts of code or text in focus, which is critical for complex projects.
The model offers two levels of reasoning enhancement: High for a balance between performance and token consumption, and Max for maximum capabilities. The API price remains at the level of the previous version GLM-5.1, which looks very attractive given the performance increase.
Numbers vs. Claims
Let's look at the key benchmarks. On Terminal-Bench 2.1, GLM-5.2 scores 81.0, closely approaching Opus 4.8 with its 85.0 and surpassing Gemini 3.1 Pro (74.0). On SWE-bench Pro — 62.1 vs. 69.2 for Opus 4.8. On FrontierSWE (long technical projects), the gap from the leader is only 1%.
However, on the DeepSWE test, GLM-5.2 shows 46.2 vs. 58.0 for Opus 4.8 and 70.0 for GPT-5.5. On NL2Repo (project generation from description), the result is 48.9 vs. 69.7 for Claude. This indicates that in some scenarios, the model still noticeably lags behind top-tier solutions.
Real-world Usage: Reviews and Criticism
Users generally positively assess the basic logic and the model's ability to work autonomously. It suggests fixes on its own and can perform complex tasks through auxiliary agents. However, there are also serious complaints.
The main problem is cloud infrastructure. Many complain about weak support and expensive pricing during peak loads. The Max mode, where the model fully unfolds, consumes significantly more tokens, making its use costly. Some developers note that the model tends to get "stuck" in infinite loops and ignore commands.
Analyst's Verdict
GLM-5.2 is undoubtedly a strong step forward. It is the best open-source model for programming today and, in a number of scenarios, truly breathes down the neck of Opus 4.8. The open MIT license and the ability to run on your own hardware are powerful advantages.
However, calling it a "killer" of Claude is premature. According to most tests, Z.ai itself places its model below Opus 4.8. And the problems with infrastructure and the high cost of using it in Max mode make it less attractive for everyday work compared to established solutions from Anthropic or OpenAI. GLM-5.2 is a formidable competitor that is closing the gap but has not yet bridged it.