A useful comparison needs exact model versions and evaluation conditions. Z.ai’s GLM-5.2 announcement names Opus 4.8 in particular tests, while Anthropic’s current documentation identifies that Opus version as legacy.
GLM-5.2 and Opus 4.8 in the June announcement
Z.ai describes GLM-5.2 as a model for long-horizon tasks with a one-million-token context and flexible effort levels. Its June post compares results on particular coding benchmarks with Opus 4.8 and other named versions. Those are the developer’s reported results, not an independent finding about all tasks or every model currently sold under the Opus name.
https://z.ai/blog/glm-5.2
Keep the competitor version explicit
Anthropic’s model page labels Opus 4.8 a legacy model and points to a newer Opus version. Specifications and lifecycle status for that exact version should not be silently substituted for a different release. The page’s context-window specification also does not by itself measure accuracy over every long-context task.
https://platform.claude.com/docs/en/models/opus-4-8/overview
A benchmark does not establish a universal UK winner
The comparison does not measure each reader’s coding workflow, cost constraints or operational requirements. Vendor benchmark results need their workload and settings, and a historical model comparison should retain its date and versions. Neither source establishes a UK-only performance advantage or one winner for every business.
https://z.ai/blog/glm-5.2
https://platform.claude.com/docs/en/models/opus-4-8/overview
Sources and further reading
https://z.ai/blog/glm-5.2
https://platform.claude.com/docs/en/models/opus-4-8/overview