Inkling
Thinking Machines Lab's first open model β a multimodal Mixture-of-Experts transformer with controllable reasoning effort
Inkling
Thinking Machines Lab β’ July 2026
Training Data
45 trillion tokens (text, image, audio, video)
Inkling
July 2026
Parameters
975B total, 41B active (MoE)
Training Method
Muon/Adam pretraining + large-scale asynchronous RL
Context Window
1,000,000 tokens
Knowledge Cutoff
2026
Key Features
Multimodal (encoder-free) β’ Controllable Thinking Effort β’ Agentic Coding β’ Open Weights
Capabilities
Multimodal: Excellent
Reasoning: Excellent
Long Context: Outstanding
What's New in This Version
First frontier open model from Mira Murati's lab; adjustable reasoning effort matches larger models at roughly one-third the token cost
Thinking Machines Lab's first open model β a multimodal Mixture-of-Experts transformer with controllable reasoning effort
What's New in This Version
First frontier open model from Mira Murati's lab; adjustable reasoning effort matches larger models at roughly one-third the token cost
Technical Specifications
Key Features
Capabilities
Other Thinking Machines Lab Models
Explore more models from Thinking Machines Lab
