As OpenAI's Former CTO Embraces 'Chinese-Style Open Source,' a Watershed Moment Arrives for AI Models

According to OpenMeter's Q2 2025 AI Model Cost Tracking Report, over the past 12 months, inference costs for mainstream large language models have dropped by an average of 67%, while the number of models supporting LoRA fine-tuning has grown by 340%. On Hugging Face, the number of PRs submitted by model contributors from China surpassed those from North America for the first time, particularly in the areas of model compression and sparsification. These figures aren't just isolated background noise; they directly point to Inkling, the newly released model from Mira Murati's startup TML—an AI model that claims to "borrow Chinese technology, focusing on balancing cost and performance."