欧洲新银行Revolut已将约2000名软件工程师从Claude Code和OpenAI Codex等专有模型迁移至由中国实验室Z.ai开发的自托管开源权重模型GLM。
Revolut Reduces AI Coding Expenses by 90% by Migrating to Chinese Open-Weight Model GLM European neobank Revolut has shifted approximately 2,000 software engineers from proprietary models like Claude Code and OpenAI Codex to a self-hosted open-weight model, GLM, developed by Chinese lab Z.ai.
Revolut AI运营负责人David Abitbol在纽约举行的AI Engineer大会上表示，此举使编程账单削减了90%，工程交付周期缩短了70%。
Speaking at the AI Engineer conference in New York, Revolut's AI operations lead David Abitbol stated the move cut coding bills by 90% and reduced engineering lead time by 70%.
为满足数据合规标准，Revolut在美国部署了约56块Nvidia B200 GPU来运行该8000亿参数模型。
Revolut operates the 800-billion-parameter model on roughly 56 Nvidia B200 GPUs hosted in the United States to meet data compliance standards.
不过，第三方分析指出，上述节省数字未计入基础设施、人员配置及巨额硬件资本支出。
However, third-party analyses note that the stated savings omit infrastructure, staffing, and substantial hardware capital expenditures.