"Inkling" Open-Weights AI Model

Inkling is a flagship, open-weight artificial intelligence model developed by Thinking Machines Lab, a startup founded by former OpenAI CTO Mira Murati. It is a massive model designed as a "balanced generalist" across reasoning, coding, and multimodal tasks. [1, 2, 3]

 Inkling: Our Open-Weights Model - Thinking Machines Lab

• Architecture & Parameters: A Mixture-of-Experts (MoE) transformer featuring 975 billion total parameters and 41 billion active parameters, supporting up to a 1M token context window.

• Multimodal Capabilities: Pretrained on 45 trillion tokens of text, images, audio, and video using an encoder-free architecture for native vision and speech processing.

• Controllable Thinking Effort: Allows developers to adjust the model's reasoning effort (from 0.2 to 0.99) to balance performance against cost and latency.

• Agentic Tool Use & Coding: Performs strongly on agentic benchmarks (such as SWEBench and Terminal Bench) and supports browser use, code generation, and multi-step tasks.

• Calibrated Epistemics: Trained using RL against proper scoring rules to improve factual accuracy, calibration (confidence estimation), and resistance to censorship without over-refusing benign prompts.

• Customization & Availability: Released with open weights on Hugging Face and integrated into the Tinker platform for fine-tuning, alongside a preview of a lighter Inkling-Small model (12B active parameters).

Mira Murati's First AI Model Is Built on China's Blueprint... Wild - YouTube

by AI Revolution

• Launch of Inkling: Thinking Machines Lab (founded by former OpenAI CTO Mira Murati) released its first AI model, Inkling, as a fully open-weights, general-purpose model.

• Architecture & China's Influence: The video analyzes how Inkling’s architecture and training methodologies draw heavily on open research and techniques pioneered by Chinese AI labs (such as DeepSeek and Moonshot's Kimi).

• Synthetic Training Data: The model was trained in part using synthetic data generated by models like Kimi, highlighting the growing trend of cross-model distillation and open-weights sharing across global AI research.

• Western vs. Chinese Open Weights: The video discusses the shift in AI dominance, contrasting Western labs' push toward proprietary closed models against China's rapid rise in high-performing open-source architectures.