Z.ai Unveils GLM-5.2: A 753-Billion-Parameter Open-Source Model for Enterprise AI

Chinese AI startup Z.ai has released GLM-5.2, a 753-billion-parameter open-source large language model (LLM) designed to excel in long-horizon coding and engineering tasks, according to the company’s official announcement. The model, available on Hugging Face, the Z.ai API, and over 20 third-party coding environments, features a 1-million-token context window and enterprise pricing starting at $12.60 per month, according to Z.ai’s documentation.

Key Features of GLM-5.2

GLM-5.2 introduces a novel architectural optimization called “IndexShare,” which reuses a single indexer across every four sparse attention layers, reducing per-token compute FLOPs by 2.9 times at the maximum 1-million-token context length. This innovation, as detailed in Z.ai’s technical whitepaper, aims to lower computational costs while maintaining performance. The model also includes an upgraded Multi-Token Prediction (MTP) layer for speculative decoding, which increases accepted token length by up to 20% during inference.

Users can toggle between “Max” and “High” thinking modes, with “Max” prioritizing logical problem-solving and “High” balancing performance with latency efficiency. According to Z.ai’s benchmark data, the “Max” mode generates nearly 85,000 output tokens per task, while “High” mode halves this output with minimal performance loss.

Benchmark Performance

On industry-standard benchmarks, GLM-5.2 outperforms many open-source models and matches proprietary leaders. In the SWE-bench Pro test, it scored 62.1, surpassing GPT-5.5 (58.6) and its predecessor GLM-5.1 (58.4). On FrontierSWE, a long-horizon task completion test, GLM-5.2 achieved 74.4%, close to Claude Opus 4.8’s 75.1%. The model also scored 77.0 on the MCP-Atlas tool-usage evaluation, outperforming GPT-5.5 (75.3), according to Z.ai’s published results.

While GLM-5.2 trails GPT-5.5 and Claude Opus 4.8 on raw Terminal-Bench 2.1 scores (81.0 vs. 85.0 and 84.0, respectively), it significantly outperforms Google’s Gemini 3.1 Pro (74.0). On the Design Arena benchmark, GLM-5.2 achieved an ELO score of 1360, beating Claude Fable 5, as reported by Z.ai.

Pricing and Availability

Z.ai offers GLM-5.2 through its GLM Coding Plan, with pricing tiers tailored to developer workflows. The Lite plan costs $12.60/month for small repositories, Pro at $50.40/month for mid-sized projects, and Max at $112.00/month for heavy workloads. API pricing is set at $1.40 per million input tokens and $4.40 per million output tokens, according to Z.ai’s pricing page.

Compared to proprietary models, GLM-5.2’s costs are significantly lower. For example, Anthropic’s Claude Opus 4.8 charges $25.00 per million output tokens, while OpenAI’s GPT-5.5 costs $30.00, as noted in a tweet by AI observer Lisan al Gaib (@scaling01). Z.ai also offers a cached input rate of $0.26 per million tokens for long-context workloads, with free cached storage during a limited-time promotion.

Developer Reception and Licensing

The model has received widespread acclaim from developers. Kilo Code confirmed day-one integration, stating, “GLM-5.2 runs in Kilo Code on day one. The 1M context window and Max effort mode are both live,” according to a post on X. Cline IDE highlighted its cost-effectiveness, noting, “GLM-5.2 is the first open-weights model to cross 80% on Terminal-Bench, and beats every other open model available.”

Z.ai’s MIT open-source license allows enterprises to use, modify, and commercialize the model without royalties or restrictive “acceptable use” policies. This contrasts with proprietary models, which often require compliance with regional restrictions and licensing fees, according to Z.ai’s technical documentation.

Implications for the AI Industry

The release of GLM-5.2 underscores the growing influence of open-source AI in enterprise settings. As U.S. proprietary models face regulatory uncertainty—such as the Trump Administration’s recent restrictions on Anthropic’s Claude Fable 5—enterprises are increasingly turning to open-source alternatives for control and cost efficiency, according to industry analysts.

“Frontier labs are absolutely scamming you on API pricing,” tweeted Lisan al Gaib, citing the disparity between open-source models like GLM-5.2 and proprietary alternatives. This sentiment reflects a broader shift toward transparency and affordability in AI infrastructure.