GLM-5.2 vs. Claude: Has the Chinese Z.ai model truly become the new king of open-source?
A new intrigue is heating up in the artificial intelligence market. The GLM-5.2 model from the Chinese company Z.ai has already been dubbed the "killer" of Anthropic's flagship Claude. But is that really the case? Let's break down the numbers and facts, setting aside the marketing noise.
GLM-5.2 is not just another update. The developers position it as a flagship model for long working sessions, capable of handling complex projects for hours without losing quality. The main technical innovation is a stable context window of 1 million tokens, five times larger than its predecessor, GLM-5.1. This means the model can "hold" an entire codebase in its memory without losing the thread of reasoning.
Key Features and Benchmarks
The model architecture offers two levels of reasoning enhancement: High for balancing performance and token consumption, and Max for maximum analysis depth. The model is distributed under the open MIT license with no regional restrictions, allowing it to be run on your own hardware (self-hosting). The API price remains at the level of the previous version.
According to Z.ai's own tests, GLM-5.2 is the strongest open model on the market. However, in most cases, it falls short of Anthropic Claude Opus 4.8. The numbers speak for themselves: on Terminal-Bench 2.1, GLM-5.2 scores 81.0 points against Opus 4.8's 85.0, but confidently surpasses Gemini 3.1 Pro with its 74.0. On SWE-bench Pro, the gap with the leader is 62.1 versus 69.2. Meanwhile, on the FrontierSWE test, where the model manages open technical projects for dozens of hours, the lag behind Opus 4.8 is only 1%.
Real Price and Pitfalls
The GLM Coding Plan subscription offers three tiers. With an annual payment and a 30% discount, the price ranges from $12.6 per month for Lite to $112 for Max. Within the subscription, quota consumption depends on load: a 3x multiplier during peak hours (2:00 PM - 6:00 PM Beijing time) and 2x off-peak. Until the end of September, a promotion makes off-peak usage billed at 1x.
Community Voice: Enthusiasm and Disappointment
Users are divided into two camps. Some call GLM-5.2 the strongest open-source model they have tried, noting that the basic logic is noticeably better than version 5.1, and in programming, it is comparable to GPT-5.5 at a high reasoning level. The model autonomously handles complex tasks, proactively suggests fixes for inconsistencies it notices, and is described as "slow, expensive, but extremely persistent in achieving its goal."
Criticism, however, focuses on service and stability. The cloud infrastructure is described as extremely weak despite a good mathematical model. Developers complain about expensive pricing and poor support, noting that it is easier to pay for Claude or GPT. The model is criticized for its tendency to get stuck in endless loops and ignore commands. According to some users, GLM-5.2 is tailored exclusively for benchmarks, not for real-world code.
Verdict: Killer or Not?
There is no clear answer. GLM-5.2 is indeed the strongest open model for programming and autonomous tasks available today. In certain long-duration scenarios, it comes very close to Anthropic's flagship. The open MIT license, the ability to run it on your own hardware, and the low entry barrier make it a notable player. However, calling it a "killer" of Claude is an exaggeration born from bloggers, not benchmarks. By most tests, Z.ai itself ranks its model below Opus 4.8. Issues with cloud infrastructure, high token consumption in Max mode, and weak support currently prevent it from becoming a mass replacement for the leaders. GLM-5.2 narrows the gap but has not yet surpassed them.
My analysis: This is a breakthrough for the open-source community, especially for developers who need a powerful local model for working with large codebases. However, for broad consumer use, the stability and ecosystem of Claude and GPT remain preferable. Keep an eye on Z.ai's development—if they solve the infrastructure problems, competition in the AI model market will become significantly fiercer.