Thinking Machines Lab unveiled Inkling, a 975-billion-parameter open-weight model designed for enterprise customization, allowing external developers to download and modify it. Inkling uses a mixture-of-experts architecture with about 41 billion active parameters per task and was trained on 45 trillion tokens across text, images, audio, and video, with multimodal ability. It is positioned as a direct alternative to closed models from OpenAI, Anthropic, and Google, aiming to compete with both proprietary and open models while emphasizing reliability by signaling uncertainty and adjusting 'thinking volume' for speed versus accuracy. The model is accessible through the Tinker platform and can be integrated via HuggingFace, supporting enterprise needs and performance benchmarking that highlights efficiency and agentic capabilities. Thinking Machines argues Inkling’s enterprise focus—enabling organizations to tailor the model to internal expertise—addresses the gap left by one-size-fits-all general-purpose models. Some benchmarks suggest Inkling leads among US open-weight models on several indexes, underscoring its position in the competitive landscape.