GLM-5.2: A Real Competitor to Claude or Just a Loud Marketing Stunt?
A major intrigue is brewing in the world of artificial intelligence. The new GLM-5.2 model from the Chinese company Z.ai has rapidly burst into the top of discussions, and many have already dubbed it the "killer" of Anthropic's flagship product, Claude. Let's figure out how justified these claims are and what this neural network actually represents.
Architecture and Key Features
GLM-5.2 is not just another update, but a flagship model designed for long and complex work sessions. Its main difference from its predecessor, GLM-5.1, is a giant leap in the size of the context window. Previously, it was 200 thousand tokens; now it is 1 million. This means the model is capable of holding in its field of view and analyzing huge volumes of code and text without loss of quality over long-term projects.
Key model parameters:
- 1 million token context: Does not degrade during ultra-long sessions, allowing work with an entire codebase in a single reasoning cycle.
- Two reasoning levels: "High" mode for balancing performance and token consumption, and "Max" mode for maximum analysis depth, requiring more resources.
- Open MIT License: Complete freedom of action — from self-hosting on your own hardware to commercial use without regional restrictions.
- API Price: The cost of requests remains at the level of the previous version, GLM-5.1, making it attractive for developers.
The model is already available on HuggingFace and ModelScope, as well as through the GLM Coding Plan subscription, the ZCode desktop agent, and the Claude Code and OpenCode environments.
Numbers and Benchmarks: What Does Real Performance Show?
According to Z.ai's own tests, GLM-5.2 is positioned as the strongest open model on the market. However, it still falls short of the leader — Anthropic Claude Opus 4.8 — in most scenarios. Let's look at the hard numbers in maximum reasoning mode:
| Benchmark | GLM-5.2 | GLM-5.1 | Opus 4.8 | GPT-5.5 | Gemini 3.1 Pro |
|---|---|---|---|---|---|
| SWE-bench Pro | 62.1 | 58.4 | 69.2 | 58.6 | 54.2 |
| Terminal-Bench 2.1 | 81.0 | 63.5 | 85.0 | 84.0 | 74.0 |
| NL2Repo | 48.9 | 42.7 | 69.7 | 50.7 | 33.4 |
| DeepSWE | 46.2 | 18.0 | 58.0 | 70.0 | 10.0 |
| ProgramBench | 63.7 | 50.9 | 71.9 | 70.8 | 39.5 |
| MCP-Atlas | 76.8 | 71.8 | 77.8 | 75.3 | 69.2 |
| Tool-Decathlon | 48.2 | 40.7 | 59.9 | 55.6 | 48.8 |
As can be seen, GLM-5.2 demonstrates impressive growth compared to its previous version, especially on the Terminal-Bench 2.1 (81.0 vs. 63.5) and DeepSWE (46.2 vs. 18.0) tests. It confidently outperforms Gemini 3.1 Pro and GPT-5.5 on a number of metrics, but still lags behind Opus 4.8. On the long-term project test FrontierSWE, the gap from the leader is only 1%, indicating colossal progress in working with long scenarios.
The Price Question: Cheap, but Not Always Stable
The GLM Coding Plan subscription offers three tiers with a 30% discount for annual payment: Lite at $12.6/month, Pro at $50.4/month, and Max at $112/month. Quota consumption depends on load: a 3x coefficient during peak hours (from 14:00 to 18:00 Beijing time) and 2x off-peak. Until the end of September, a promotion is active where off-peak usage is billed at 1x.
Users note that the model only reveals itself in Max mode, which consumes significantly more tokens. At the same time, the cloud infrastructure is criticized as extremely weak, and support as insufficient. Many developers complain that it's easier to pay for Claude or GPT than to put up with the instability of Z.ai's service.
Real Reviews: Enthusiasm vs. Disappointment
User opinions are divided. The model's strengths:
- The best open model for programming and autonomous tasks at the moment.
- Basic logic is noticeably improved compared to version 5.1.
- Ability to independently perform complex tasks through auxiliary agents and suggest fixes.
- Persistence in achieving a goal, despite slowness and high cost.
Criticism primarily concerns the service and stability:
- Weak cloud infrastructure despite a good mathematical model.
- High billing cost and weak support.
- Tendency to get stuck in infinite loops and ignore user commands.
- Suspicion that the model is "tuned" exclusively for benchmarks.
Analyst's Verdict: Killer or Not?
There is no clear answer. GLM-5.2 is undoubtedly the strongest open model today. It has come very close to Anthropic's flagship in certain scenarios, especially in long-term projects. The open MIT license, the possibility of self-hosting, and the low entry barrier make it a notable player in the market.
However, calling it a "Claude killer" is premature. According to most tests, Z.ai itself places its model below Opus 4.8. Moreover, the raw and unstable infrastructure, high token consumption, and weak support negate many of its advantages. For now, GLM-5.2 is a promising but unfinished product that narrows the gap with the leaders but does not surpass them.
My professional opinion: GLM-5.2 is an important step for the open AI model industry, demonstrating that competition with closed giants is possible. However, for mass adoption, Z.ai's developers will need to solve the infrastructure and stability issues. For now, it is a tool for enthusiasts and those willing to put up with imperfections for access to a cutting-edge open architecture.