Z.AI’s missing Air model frustrates community as GLM 5.1 proves too large for local use

A vocal segment of the Z.AI (Zhipu AI) community has expressed frustration over the company’s recent model releases. The last Air model — Z.AI’s compact, easily deployable variant — shipped at version 4.5, and no upgraded successor has appeared since. Meanwhile, GLM 4.7 Turbo arrived as a strong contender, but was quickly overtaken by the Qwen 3.6 35B model for agentic coding tasks, which offers superior performance with fewer tokens. Now GLM 5.1 has been released as a coding beast, but its massive size makes it impractical for most local hardware, and even API inference is sluggish. Users are left wondering: when will Z.AI bring back the Air line with frontier-level reasoning and knowledge, or deliver a turbo model that can outpace Qwen 3.6 in agentic coding while keeping token counts low?

Beyond raw size concerns, the community is also pushing for Z.AI to adopt quantization-aware training (QAT), similar to what Google did with Gemma. Such an approach would allow Z.AI to produce models that are both powerful and efficient at inference, leaving Qwen in the dust. The underlying sentiment is clear: Z.AI’s recent trajectory favors brute-force scale over practicality, and developers who need local deployment or fast API responses are being left behind. Without a compact, high-performance Air or turbo model, Z.AI risks ceding the accessible frontier to competitors like Qwen and Gemma.