GLM-5.2 is the latest flagship model from Zhipu AI, the Chinese lab known overseas as Z.ai, and it is one of the most important open-weight releases of 2026. Shipped on June 13, 2026, it is a coding-first model that is openly available under a permissive license, carries a 1 million token context window, and, by several independent measures, ranks as the most capable openly available model on the market. In plain terms, a Chinese company has released a model that competes with the top Western systems on hard coding tasks, put its weights out for anyone to download, and priced its API at a fraction of the closed alternatives.
The short version: GLM-5.2 is a large, coding-focused model released under the MIT open-source license, which means anyone can download, run, modify, and commercialize it without royalties or usage restrictions. It beats GPT-5.5 on several long-horizon coding benchmarks, lands close to Claude Opus 4.8, and costs several times less to use through its API. This piece explains what GLM-5.2 is, who makes it, how it performs, what its open license and pricing mean, and why it matters. For context on the wider open-weight field, see our coverage of Llama and DeepSeek.
What GLM-5.2 is
GLM-5.2 is the newest model in Zhipu’s GLM family, and it is built specifically for coding and long-horizon engineering work, the kind of multi-step, autonomous tasks where a model has to plan, write, and revise code over a long session. It is a mixture-of-experts model with roughly 750 billion total parameters, of which about 40 billion are active for any given token. That design lets it hold the knowledge of a very large model while doing the work of a much smaller one on each step, which keeps it efficient to run relative to its size.
Its headline specification is the 1 million token context window, a large jump from the 200,000 tokens of the previous GLM-5.1. A context that large lets the model hold entire codebases, long documents, or extended histories in a single request, which is exactly what long-horizon coding needs. The model is available on Hugging Face, through the Z.ai API, and across more than 20 third-party coding environments, so it is easy to reach whether you want to run the weights yourself or call a hosted endpoint.
Who makes it: Zhipu AI (Z.ai)
Zhipu AI is a major Chinese AI lab and the company behind the GLM series, which stretches back through GLM-4 in 2024 to the current GLM-5 line. Overseas it operates as Z.ai, and it is now a public company, having listed in Hong Kong in early 2026. GLM-5.2 is the model that has drawn the most attention to the company yet, both for its capability and for the strategy behind releasing it openly, and its progress is part of a broader story of Chinese labs closing the gap with US frontier developers.
How GLM-5.2 performs
The benchmarks are the reason GLM-5.2 made news, and they are strongest exactly where the model is aimed: real, long-horizon coding. On SWE-bench Pro, a demanding software-engineering benchmark, GLM-5.2 scored 62.1, ahead of GPT-5.5 at 58.6 and its own predecessor at 58.4. On other coding-focused evaluations it lands right in the mix with the best closed models: around 74% on FrontierSWE, essentially tied with Claude Opus 4.8, and a close second to Opus on the MCP-Atlas tool-use benchmark. Independent testing summarized it as beating GPT-5.5 on multiple long-horizon coding benchmarks for roughly a sixth of the cost.
Two caveats keep this honest. First, GLM-5.2 is coding-first, so these results say more about software engineering than about every kind of task, and a model tuned for code will not automatically lead everywhere. Second, benchmarks are a guide, not a guarantee; the real test is how the model performs on your own work. But the direction is clear: on hard coding, an open-weight model is now trading blows with the top closed systems.
What the open license and pricing mean
The part that matters most strategically is the license. Zhipu released GLM-5.2’s weights under the MIT license, one of the most permissive open-source licenses there is. That makes it what some call "pure open": you can download the model, run it on your own hardware, modify it, and build commercial products on it without paying royalties or agreeing to the restrictive acceptable-use terms attached to many so-called open models. For a business that wants control over where its models run and how they are used, that is a meaningful difference from a closed API or a dual-use license.
Pricing reinforces the point. Through the Z.ai API, GLM-5.2 costs about $1.40 per million input tokens and $4.40 per million output tokens, roughly three to seven times cheaper than leading closed models, with subscription tiers starting near $13 a month. Between the open weights and the low API price, GLM-5.2 is aggressively positioned to win developers on cost and control, not just capability.
Why it matters
GLM-5.2 is a marker of how fast the open-weight frontier is moving. A capable, openly licensed coding model that competes with the best closed systems changes the calculus for teams that assumed frontier-level coding meant a closed API and a premium bill. It gives developers a high-end option they can self-host, audit, and build on freely, and it intensifies the competition that has been pushing prices down and capabilities up across the field. It also underscores that the strongest open models are increasingly coming from Chinese labs, alongside DeepSeek and others, which is a notable shift in where open AI progress originates.
Frequently Asked Questions
What is GLM-5.2?
GLM-5.2 is the latest flagship model from Zhipu AI (Z.ai), released June 13, 2026. It is a coding-first, mixture-of-experts model with roughly 750 billion total parameters and about 40 billion active per token, a 1 million token context window, and openly available weights under the MIT license. It ranks among the most capable openly available models, especially for software engineering.
Who makes GLM-5.2?
Zhipu AI, a major Chinese AI lab known overseas as Z.ai, makes GLM-5.2. It is the newest entry in the company’s GLM model family and was released after Zhipu listed publicly in Hong Kong in early 2026. Zhipu is one of the Chinese labs closing the gap with leading US AI developers.
Is GLM-5.2 really open source?
Its weights are released under the MIT license, one of the most permissive open-source licenses, so you can download, run, modify, and commercialize the model without royalties or restrictive acceptable-use terms. That is more open than many models marketed as open, which attach usage restrictions. You still need capable hardware to run a model this large yourself.
How good is GLM-5.2 at coding?
Very good by current benchmarks. It scored 62.1 on SWE-bench Pro, ahead of GPT-5.5, and lands near Claude Opus 4.8 on other coding and tool-use benchmarks such as FrontierSWE and MCP-Atlas. It is built for long-horizon, multi-step engineering tasks. As always, benchmarks are a guide, so the real measure is how it does on your own work.
How much does GLM-5.2 cost?
Through the Z.ai API, GLM-5.2 costs about $1.40 per million input tokens and $4.40 per million output tokens, roughly three to seven times cheaper than leading closed models, with subscription tiers starting around $13 a month. Because the weights are open, you can also run it on your own hardware instead of paying per token.
How does GLM-5.2 compare to closed models like GPT-5.5 and Claude Opus 4.8?
On several long-horizon coding benchmarks GLM-5.2 beats GPT-5.5 and is roughly level with Claude Opus 4.8, at a fraction of the price and with open weights. It is strongest on coding specifically, so the comparison is closest there. For broader tasks, the leading closed models may still have an edge, but the gap on hard coding has narrowed sharply.
What is the context window on GLM-5.2?
GLM-5.2 has a 1 million token context window, a large increase from the 200,000 tokens of GLM-5.1. That lets it hold entire codebases or long document sets in a single request, which is important for the long-horizon coding and engineering work it is designed for.
