Crypto news

In recent days, the crypto community and AI enthusiasts have been actively discussing the new GLM-5.2 model from the Chinese company Z.ai. It has already been dubbed the "killer" of Anthropic's Claude, and as is often the case, there is more hype here than real substance. Let's figure out what this model is and how truly dangerous it is for market leaders.

What is GLM-5.2 and what are its main advantages?

GLM-5.2 is a flagship open-source model that, according to its developers, is designed for handling long and complex work sessions. The key difference from its predecessor GLM-5.1 is a stable context window of 1 million tokens (compared to 200 thousand previously). This means the model can keep a huge volume of code or text in its "field of view" without losing quality.

Key features that have attracted attention:

• 1 million token context that does not degrade during ultra-long sessions.
• Two levels of reasoning enhancement: High (balance of performance and token consumption) and Max (maximum capabilities, but with high resource consumption).
• Open MIT license with no regional restrictions, allowing the model to be run on your own hardware (self-hosting).
• API price remains at the level of the previous version, which is an important factor.

The model is available on HuggingFace and ModelScope, as well as through the GLM Coding Plan subscription, the ZCode desktop agent, and even Claude Code and OpenCode environments.

Benchmarks: Where is GLM-5.2 strong, and where is it weak?

According to Z.ai's own tests, GLM-5.2 is recognized as the strongest open-source model on the market. However, it falls short of Anthropic's flagship — Claude Opus 4.8 — in most scenarios.

On standard programming tests, the gap from GLM-5.1 is noticeable: 81.0 vs 63.5 on Terminal-Bench 2.1 and 62.1 vs 58.4 on SWE-bench Pro. Meanwhile, on Terminal-Bench 2.1, the score of 81.0 closely approaches Opus 4.8 (85.0) and surpasses Gemini 3.1 Pro (74.0).

Comparison table in Max mode (key tests):

On long-horizon tasks, the picture is similar. On the FrontierSWE test, GLM-5.2 lags behind Opus 4.8 by only 1%, but surpasses GPT-5.5 and the previous version Opus 4.7. On PostTrainBench, the model outperforms Opus 4.7 and GPT-5.5, yielding only to Opus 4.8. However, on the ultra-long SWE-Marathon, the gap from Opus 4.8 is already 13%.

The cost factor and real reviews

The GLM Coding Plan subscription is divided into three tiers: Lite ($12.6/month), Pro ($50.4/month), and Max ($112/month). Quota consumption depends on load: a 3x multiplier during peak hours and 2x off-peak. Until the end of September, a promotion is active where off-peak usage is billed at 1x.

Users are divided in their opinions. Strengths: the model is praised for better base logic compared to 5.1, comparability with GPT-5.5 at a high reasoning level, and the ability to autonomously perform complex tasks. However, it is criticized for weak cloud infrastructure, high token consumption in Max mode, and a tendency to get stuck in infinite loops. Many note that the model only reveals its potential in Max mode, which consumes resources many times more than High.

Conclusion: Killer or not?

There is no clear answer. GLM-5.2 is the best open-source model today for programming and autonomous tasks. In certain scenarios, it comes very close to Anthropic's flagship. The open MIT license, the ability to run on your own hardware, and the low entry barrier make it a notable player.

My professional opinion: calling GLM-5.2 a "killer" of Claude is more of a marketing move than reality. In most tests, Z.ai itself ranks its model below Opus 4.8. Moreover, users complain about unstable infrastructure and high token consumption. The model is closing the gap with the leaders, but it hasn't overtaken them yet. For enthusiasts and developers who need a powerful open-source model, this is an excellent choice. But for those seeking stability and predictability, Claude or GPT remain more reliable options.