Revolut says it has cut its coding costs by 90% using AI after shifting much of its work from OpenAI and Anthropic to GLM-5.3, an open AI model developed by Chinese company Z.ai. The fintech now runs this model on 56 NVIDIA B200 GPUs, serving approximately 2,000 engineers. According to its head of AI operations, costs have fallen to around $0.30 per million tokens. The company also reports a 60% improvement in coding times and a 70% reduction in the time needed to deploy software changes. Although Revolut continues to use Claude Code, Codex, and Gemini for certain tasks, its own GLM-based system now handles much of the work. If more companies adopt this strategy, OpenAI and Anthropic could face considerable pressure on their pricing and business models.