Z.ai Releases Open-Weights GLM-5.2 LLM, Outperforms GPT-5.5 in Coding Benchmarks
Chinese AI startup Z.ai has launched GLM-5.2, a 753-billion parameter open-weights large language model (LLM) designed for long-horizon autonomous coding and engineering tasks. The model, available with a 1-million-token context window, achieved higher scores than OpenAI's GPT-5.5 on multiple industry-standard coding benchmarks, including SWE-bench Pro and FrontierSWE. GLM-5.2 is released under an unrestricted MIT open-source license, allowing enterprises to download, customize, and run it locally, offering a cost-effective alternative to proprietary models. Its API pricing is also significantly lower than some Western rivals.
Chinese AI startup Z.ai (formerly Zhipu AI) has announced the immediate release of GLM-5.2, a 753-billion parameter open-weights large language model (LLM). This model is engineered to excel in "long-horizon" autonomous coding and engineering tasks.
GLM-5.2 is available on Hugging Face, the Z.ai API, and over 20 third-party coding environments. It features a stable 1-million-token context window and enterprise subscription tiers starting at $12.60 per month. The model's core weights are released under an unrestricted MIT open-source license, enabling businesses to download, customize, and run it locally or via virtual machines, incurring only compute and electricity costs. This open-source approach offers an alternative to proprietary models, especially amidst regulatory uncertainties affecting some state-of-the-art models.
Architecturally, GLM-5.2 includes an optimization called "IndexShare," which reuses a single indexer across every four sparse attention layers. This innovation reportedly reduces per-token compute FLOPs by 2.9 times at its maximum 1-million-token context length. The model also features an upgraded Multi-Token Prediction (MTP) layer, which can boost accepted token length by up to 20% during inference. Users can also select between "Max" and "High" "Thinking Modes" to balance problem-solving effort with token efficiency.
On industry-standard third-party benchmark tests, GLM-5.2 performed strongly, often matching or surpassing closed-weights rivals like OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.8. Specifically:
* **SWE-bench Pro:** GLM-5.2 scored 62.1, outperforming GPT-5.5 (58.6) and its predecessor, GLM-5.1 (58.4). * **FrontierSWE (Dominance):** It achieved 74.4%, surpassing GPT-5.5 (72.6%) and nearing Claude Opus 4.8 (75.1%). * **MCP-Atlas:** GLM-5.2 scored 77.0, higher than GPT-5.5 (75.3). * **Humanity's Last Exam (w/ Tools):** With tools, it reached 54.7, ahead of GPT-5.5 (52.2). * **PostTrainBench & SWE-Marathon:** In multi-hour engineering workloads, GLM-5.2 consistently topped GPT-5.5, scoring 34.3% vs. 25.0% on PostTrainBench and 13.0% vs. 12.0% on SWE-Marathon. It also secured first place on the crowdsourced Design Arena benchmark, beating Claude Fable 5.
While GLM-5.2's Terminal-Bench 2.1 score of 81.0 trailed Claude Opus 4.8 and GPT-5.5 slightly, it significantly outscored Google's Gemini 3.1 Pro (74.0). The model's "Max" thinking mode pushed performance but used more output tokens, while "High" mode balanced performance with token efficiency.
Z.ai also launched the GLM Coding Plan, with tiers ranging from $12.60 to $112.00 per month (billed annually), offering support for third-party coding harnesses. The GLM-5.2 API is priced at $1.40 per million input tokens and $4.40 per million output tokens. This pricing is substantially lower than some leading Western proprietary models; for instance, its output token cost is about one-sixth of GPT-5.5's.
The release has received positive feedback from the developer community, with Kilo Code, Cline IDE, and Eigent AI confirming day-one integrations and highlighting its capabilities and cost-effectiveness. According to VentureBeat, observers have noted the significant pricing difference between open-weights innovators and proprietary labs.
(Source: VentureBeat)
