GLM-5.2 from Z.ai: Has This Chinese Model Really Surpassed Claude? Analyzing the Hype
In recent days, the crypto and tech community has been buzzing with discussions around a new open-source neural network — GLM-5.2 from Z.ai. Many have been quick to dub it the "killer" of Anthropic's flagship model, Claude Opus 4.8, citing impressive benchmarks at a significantly lower price. Let's examine how justified such bold claims are and what this new model actually represents.
GLM-5.2 is a flagship model optimized for long working sessions and autonomous execution of complex tasks. Its main difference from its predecessor GLM-5.1 is an expanded context window of up to 1 million tokens, allowing it to process entire codebases in a single reasoning cycle. Key features include: two levels of reasoning depth (High for balance and Max for maximum performance), an open MIT license allowing the model to be run on your own hardware, and API access at a price comparable to the previous version.
Numbers and Benchmarks: What Do the Tests Show?
According to Z.ai's internal tests, GLM-5.2 indeed demonstrates a significant performance increase compared to GLM-5.1. On the Terminal-Bench 2.1 test, it scored 81.0 points versus 63.5, and on SWE-bench Pro, 62.1 versus 58.4. However, it still lags far behind the leader, Claude Opus 4.8 (85.0 and 69.2 points respectively). At the same time, GLM-5.2 confidently outperforms Gemini 3.1 Pro and GPT-5.5 on a number of key metrics, especially in tasks related to command-line work and autonomous use of external tools via the MCP protocol.
On long-horizon tasks such as FrontierSWE, GLM-5.2 trails Opus 4.8 by only 1%, which is an impressive result. However, on more complex tests like SWE-Marathon, the gap widens to 13%. Thus, the model confidently holds the position of the best among open-source alternatives, but it lacks stability in the most challenging scenarios to achieve absolute leadership.
Pricing and Real User Experience
The GLM Coding Plan subscription offers three tiers: Lite ($12.6/month), Pro ($50.4/month), and Max ($112/month) with annual billing. This is significantly cheaper than subscriptions to Claude or GPT, which has attracted the attention of enthusiasts. However, user experience has been mixed. The model's strengths include excellent basic logic, the ability to autonomously correct errors, and persistence in achieving goals. Criticism mainly concerns the cloud infrastructure: users complain about instability, high token consumption in Max mode, and weak support. Many note that the model only "shines" in the resource-intensive Max mode, which negates its price advantage.
Analyst's Verdict
GLM-5.2 is undoubtedly a strong player in the open-source AI model market. It demonstrates impressive results in programming and autonomous tasks, narrowing the gap with industry leaders. The open license and self-hosting capability make it attractive to developers. However, calling it a "killer" of Claude is premature. On most benchmarks, it falls short of Opus 4.8, and infrastructure issues along with high token consumption in maximum mode reduce its practical value.
My expert opinion: GLM-5.2 is an excellent demonstration of progress, but not a revolution. The AI model market is moving towards multimodality and specialization, and Z.ai has potential here. However, until the user experience reaches the level of stability and predictability seen with Anthropic or OpenAI, it's too early to talk about a change in leadership. It's a benchmark champion, but not always the best tool for real daily work.