GLM 5.2 Becomes Base44's Default Superagent Model

Base44 Swaps Its Default Agent Model

Base44 changed the default model behind Superagent conversations, moving from Anthropic's Sonnet 4.6 to GLM 5.2, an MIT-licensed open-weight model from Z.ai built for coding and long-horizon agentic work. Sonnet 4.6 was retired from the model picker entirely as part of the change.

What Changed and What Didn't

Any Superagent left on its default configuration now runs on GLM 5.2 for standard conversational turns. Two important carve-outs remain in place: image and PDF-based turns continue to run on Sonnet 5 regardless of a workspace's chosen default, since GLM 5.2 handles multimodal input less consistently. And Enterprise workspaces are not affected by this default change, Enterprise plans continue defaulting to Sonnet 5, including for the build-mode planner used in more structured, multi-step build workflows.

Why This Is a Significant Change

For a vibe-coding platform, the model sitting behind the default agent experience is one of the most consequential product decisions Base44 can make, it directly shapes response quality, latency, and cost for the majority of everyday users who never touch the model picker. GLM 5.2 is notable in its own right: it's recognized as a top-performing frontier-class coding model with a 1M-token context window and multiple reasoning-effort modes, and it is fully open-weight, which likely gives Base44 more flexibility on hosting economics than a closed frontier model would. Swapping the default, not just adding an option, signals real confidence in GLM 5.2's quality for the bulk of Superagent traffic.