@fedor @felipecsl many open models still get really dumb at >half context full. GLM 5.3 Flash and Deepseek 4.1 Flash i’ve both noticed being rubes, forgetting all skills, etc. harness: codex
cited on: deepseek-v4-1-flash · glm-5-3-flash
Reproduced against link rot, credited and linked to its original. Yours and you’d rather it weren’t here? Open an issue.