Claude Fable 5 across five agent workloads vs glm-5.2, opus-4-8 and sonnet-5: mid-tool-call refusals, adaptive thinking, and cost that shifts 5-15x by shape.

Claude Fable 5 across five agent workloads vs glm-5.2, opus-4-8 and sonnet-5: mid-tool-call refusals, adaptive thinking, and cost that shifts 5-15x by shape.

I ran the same real agent task on GLM 5.2 and Claude Fable 5. Fable 5 is faster and stronger on benchmarks, but GLM 5.2 finished for $2.76 vs over $10. The honest cost math for…