- GLM-5.2, a 753-billion-parameter open-weights model released June 16, beats OpenAI's GPT-5.5 on multiple coding benchmarks while costing roughly one-sixth as much to operate, marking a significant milestone for open-source AI development.
- On the SWE-bench Pro benchmark, which tests real-world software engineering tasks, GLM-5.2 scored 62.1, decisively outperforming GPT-5.5’s 58.6.
- On FrontierSWE, a test designed to measure long-horizon task completion, GLM-5.2 achieved 74.4 percent, surpassing GPT-5.5 at 72.6 percent.
- The cost advantage is striking; GLM-5.2 API pricing runs $5.80 per million tokens, while GPT-5.5 costs $35 per million tokens, making GLM-5.2 roughly one-sixth the price.
- GLM-5.2 operates with 753 billion parameters and introduces a major architectural optimization called IndexShare, which reuses the same indexer across every four sparse attention layers.
- This single innovation reduces per-token compute operations by 2.9 times at the maximum 1-million-token context length.
- GLM-5.2 features an upgraded Multi-Token Prediction layer for speculative decoding that boosts accepted token length by up to 20 percent during inference.
- Z.ai released GLM-5.2’s weights under an unrestricted MIT open-source license, allowing enterprises to download the model freely from Hugging Face.
- This licensing approach contrasts with many dual-use licenses that impose restrictive governance policies.
Z.ai's GLM-5.2, released on June 16, is a groundbreaking 753-billion-parameter open-source model that has outperformed OpenAI's GPT-5.5 on multiple coding benchmarks while being significantly cheaper to operate.1
On the SWE-bench Pro benchmark, which evaluates real-world software engineering tasks, GLM-5.2 scored 62.1, surpassing GPT-5.5's 58.6, according to VentureBeat.2
In another test, FrontierSWE, designed to measure long-horizon task completion, GLM-5.2 achieved 74.4%, outperforming GPT-5.5's 72.6% and closely trailing Anthropic's Claude Opus 4.8 at 75.1%.3
The cost advantage is striking: GLM-5.2's API pricing is $5.80 per million tokens, while GPT-5.5 costs $35 for the same amount, making GLM-5.2 roughly one-sixth the price for comparable or superior performance.4
The model's architecture includes a major optimization called IndexShare, which reduces per-token compute operations by 2.9 times at a maximum context length of 1 million tokens.56
GLM-5.2 is available under an unrestricted MIT open-source license, allowing enterprises to download, customize, and run it locally or on virtual machines for minimal costs.8
Additionally, it scored 81.0 on the Terminal-Bench, making it the first open-weights model to cross 80% on that benchmark, and it also excelled in other tests, scoring 77.0 on MCP-Atlas and 54.7 on Humanity's Last Exam.
“Z.ai's GLM-5.2, a 753-billion-parameter model, outperforms GPT-5.5 on several benchmarks, including SWE-bench Pro and FrontierSWE. The model's operational cost is significantly lower, making it an attractive option for enterprises.”
eciks.org