GLM-4.5: Reasoning, Coding, and Agentic Abililties (Z.ai official blog)

Read https://z.ai/blog/glm-4.5 for the benchmark figures on models/glm-4-5. As recorded by the reader: Values read from the official GLM-4.5 tech report (Tables 4 and 6) and independently matched on the Z.ai release blog. MMLU 90.0 and IFEval 86.1 come from the general-abilities Table 6; MMLU-Pro 84.6, AIME 24 91.0, MATH 500 98.2 and SWE-bench Verified 64.2 appear in both sources (AIME Avg@32, GPQA Avg@8). Caveat: the source labels the row simply 'GPQA' (79.1); it is reported under the closest allowed name 'GPQA Diamond'. Two other published rows do not match any allowed exact string and are therefore omitted: LiveCodeBench 2407-2501 (72.9) and Terminal-Bench (37.5, version unspecified, Terminus harness).