note: InferenceBench was released only a month ago, so we know that GLM-5.2 couldn't have been optimized on InferenceBench. yet, its results are very strong. this serves as additional evidence that it's indeed a good model, comparable to proprietary frontier models.
you can download and play around with the traces here: https://huggingface.co/datasets/aisa-group/InferenceBench-Trajectories
or you can view the traces directly here: https://inferencebench.ai/
💥NEW: important updates on InferenceBench: - GLM-5.2 (Max) results are very strong (7.0x speed-up compared to Opus 4.8's 7.6x speed-up), - we have a trace viewer now available on our website, - all our traces are hosted on HuggingFace.
