Thinking Machines Releases Inkling AI Model

Thinking Machines Lab announced the release of Inkling, their 975B-parameter multimodal open-weights mixture-of-experts (MoE) AI model with 41B active parameters, designed for customized applications and controllable thinking efforts.

Inkling introduces a new level of customization and efficiency in AI models by offering controllable thinking effort, potentially reducing token usage and improving performance on various benchmarks.

Developers, AI researchers, and organizations seeking adaptable AI models for specific use cases could leverage Inkling's open weights and fine-tuning capabilities.

While Inkling shows promise, issues such as high VRAM requirements, role-play risks, and lower performance on some benchmarks compared to competitors highlight potential adoption challenges.