GPT-6 Luna
GPT-6 Luna's own checks, newest first. Its results on the tests every model ran are on the Quick suite and Deep suite pages.
Medium against xhigh on the deep suite, 8 to 10 November 2026#
GPT-6 Luna by kind of task
On big changes, xhigh solved 10 of 30 tries against 0 at medium.
Source: WorkloadBench at workloadbench.brenhq.com, updated 10 November 2026. Deep suite, 182 of GPT-6 Luna's runs at medium, high and xhigh effort.
Notes
Big changes are the 6 tasks with the largest reference changes in the suite, small fixes the 8 with the smallest. Each bar counts tries: 5 runs on each task of the group, 3 at high.