Zhipu AI's GLM-5.2 Nears Top Closed-Source Models in Coding Benchmarks
Chinese artificial intelligence laboratory Zhipu AI has launched its new GLM-5.2 model, an open-source offering available under the MIT license. This model features a stable 1-million-token context window. On the FrontierSWE benchmark, which evaluates hours-long coding tasks, GLM-5.2 demonstrated performance close to leading closed-source models, trailing Anthropic's Claude Opus 4.8 by only one percentage point. However, its reasoning capabilities are reported to be significantly behind those of its closed-source competitors.
Chinese artificial intelligence laboratory Zhipu AI has announced the release of its GLM-5.2 model. This new open-source model is distributed under the MIT license and boasts a stable 1-million-token context window.
The GLM-5.2 model has shown strong performance in specific benchmarks, particularly in coding tasks. On FrontierSWE, a benchmark designed to evaluate performance in hours-long coding challenges, GLM-5.2 performed closely to industry leaders. It trailed Anthropic's Claude Opus 4.8 by a margin of just one percentage point.
Despite its competitive showing in coding, the model's performance in reasoning tasks indicates room for improvement. GLM-5.2 reportedly falls well behind its closed-source rivals in this area.
(Source: The Decoder AI)
