LisanBench model rankings shift without the edit-distance-one rule

In an October 24 LisanBench update, a tester said Meta's Muse Wonder Full and Muse Wonder took the top two scores at maximum reasoning effort but were slightly less efficient than Claude Opus 5.5. Removing the edit-distance-one rule put GPT-6 Astra medium first, followed by Gemini 4 Argon high. Meta provided free access to the Muse models; the tester said the tests and conclusions were their own.