Latest release
v 0.10.0
Changed 4 Blog 1
Changed 4 entries
- GLM template moved to a three-tier model mapping: Opus slot now uses
glm-5.2[1m], sonnet usesglm-5.1, haiku usesglm-5-turbo. The previous template put GLM-5.1 in opus and reused GLM-5-Turbo for both sonnet and haiku, so there was no real middle tier. - GLM-5.2 is the new opus-tier model with a 1,000,000-token context window, exposed through the
[1m]model-name suffix. GLM-5.1 and GLM-5-Turbo are 200K models. - Removed the global
CLAUDE_CODE_AUTO_COMPACT_WINDOW(131072) andCLAUDE_AUTOCOMPACT_PCT_OVERRIDEoverrides from the GLM template. A single global window cannot fit a 1M model and two 200K models at once, so each tier now carries its own size: the[1m]suffix for GLM-5.2, and Claude Code's default 200K assumption for the other two, which matches their real limit. - Updated GLM model and context-window references across the provider docs, FAQ, environment-variables guide, and glossary to reflect the three models.
Blog 1 entry