Databricks, Microsoft embrace Chinese open-source AI models

Zhipu AI's GLM-5.2, a 744-billion-parameter open-weight model released under the MIT license by Chinese startup Z.ai, is drawing growing adoption from American technology companies looking for cheaper, self-hosted alternatives to proprietary AI systems. Databricks and other enterprises have begun integrating the model into coding and agent workflows, marking a new phase in the competition between open-source Chinese models and closed Western incumbents.

Enterprise Appeal

Zoho founder Sridhar Vembu noted on X on June 29 that Databricks is embracing GLM-5.2, adding that Microsoft ↘ 1.80% had earlier signaled interest in DeepSeek. The model's appeal to enterprises rests on concrete economics: Z.ai's Coding Plan starts at roughly $10 per month, a fraction of the cost of Anthropic's Claude Code tiers, and the MIT-licensed weights can be downloaded and deployed on private infrastructure with no provider-side oversight.aiweekly+4

Databricks' Yuchen Jin described GLM-5.2 as the point at which open-source models deliver "Claude-level performance for coding and agent workflows," enabling companies to download weights they can fine-tune and retain internally.digg

Benchmark Performance

GLM-5.2 scored 62.1 on SWE-bench Pro, surpassing GPT-5.5's 58.6, and ranks as the top open-weights model on Artificial Analysis's Intelligence Index. Independent benchmarker Artificial Analysis places it fourth overall globally, behind Claude Fable 5, Opus 4.8, and GPT-5.5. The model features a one-million-token context window and 40 billion active parameters in a mixture-of-experts architecture.ofox+2

Former Meta Platforms ↘ 3.27% and Google Alphabet Inc. ↘ 1.68% DeepMind vice president Matt Velloso called it "the first open model that passes the bar as a daily driver," saying it "doesn't talk too much, doesn't go in circles trying to explain itself, just does the job".aiweekly

Security Concerns

The model's openness has raised national security alarms. The Verge reported that researchers found GLM-5.2 matches Anthropic's restricted Mythos model in certain bug-finding and vulnerability-discovery scenarios. Because the weights are freely downloadable, the model can be run by anyone without oversight — a capability the Trump administration views as a threat, given its efforts to restrict China's access to advanced AI.theverge+1

The dual-use tension is now central to the policy debate: the same qualities that make GLM-5.2 attractive to cost-conscious enterprises — openness, local deployment, no licensing barriers — also make it accessible to malicious actors seeking to exploit software vulnerabilities at scale.theverge