Z.ai's GLM-5.2 beats Llama, costs 10x less than closed rivals

Zhipu AI's GLM-5.2, released June 13, is a game-changing open-weight model. It achieves top scores on coding and long-horizon agentic benchmarks among open models, and trails only slightly behind closed frontier models like Fable 5, Opus 4.8, and GPT-5.5. Its API costs roughly one-tenth of Anthropic or OpenAI's best offerings, and the model can be downloaded and run locally for free. Coinbase CEO Brian Armstrong is shifting architecture to let developers choose models, implicitly endorsing GLM-5.2 as a cost-saving alternative. This disrupts the U.S. closed-model duopoly.

Behind GLM-5.2 lies a broader trend: Chinese labs have collectively released more MIT/Apache 2.0-licensed open-weight models in 2026 than all non-Chinese labs combined. Eight labs—including DeepSeek, Alibaba's Qwen, Moonshot's Kimi, Xiaomi's Mimo, and Zhipu—drive this, with six models now appearing on major capability rankings. The U.S. has largely shunned open-source, relying on Meta's Llama, but Llama's best model cannot match GLM-5.2. The author argues the U.S. must pivot to supporting open-weight development to remain competitive, as Chinese models already match or exceed closed alternatives at a fraction of the cost.