Stocks● Neutral
GLM-5.2: Tsinghua Lab's Open-Source Model Undercuts ChatGPT
Tsinghua's GLM-5.2 matches ChatGPT and Claude on benchmarks, costs 5-7x less, and offers a million-token context window under an MIT license.
Continue with
A Chinese university lab just built one of the best AI models of 2026, matching ChatGPT and Claude at a fraction of the price.
Tsinghua University started a research project in 2021. Five years later, the lab created GLM-5.2, an open-source model that independent benchmarks now call the best open-source AI of the year. In coding tasks, it edges past both Claude and ChatGPT, according to published benchmark scores.
GLM, or General Language Model, launched through Z.ai in June 2026. It targets coding and agentic work – multi-step tasks where the model plans, executes, and iterates without human intervention. The newcomer entered a closed-door race and immediately outperformed its rivals.
Why the One-Million-Token Context Window Matters
One million tokens equals roughly 750,000 words, or about ten novels. Every AI has a limited context window – a word limit within which it retains memory. For most models like ChatGPT and Claude, that limit sits around 100,000 tokens. GLM-5.2's window is ten times larger.
A software developer can paste an entire codebase into GLM-5.2 and the model reasons across all of it at once, without forgetting earlier sections. That is a practical leap for debugging, refactoring, and large-scale code review.
MIT License: What Open-Source AI Looks Like in 2026