Flow - 4x faster than typing
Mini toolsInkling
Inkling is a Mixture-of-Experts transformer with 975B total parameters and 41B active per token, routing to 6 of 256 experts plus 2 shared experts via a sigmoid-based router. It supports a 1M-token context window and accepts text, image, and audio inputs, producing text outputs through an encoder-free multimodal design. Inkling offers controllable thinking effort, letting developers tune the balance between reasoning depth and token cost. It was pretrained on 45 trillion tokens of text, image, audio, and video, then post-trained with large-scale reinforcement learning across math, agentic coding, tool use, multimodal, chat, and safety domains. Released with full open weights under Apache 2.0, Inkling is intended as a customizable foundation model for fine-tuning, supporting agentic tool use, coding assistants, and general-purpose multimodal applications.
New
Multimodal
Overview
Inkling is an open-weights Mixture-of-Experts transformer from Thinking Machines Lab with 975B total parameters (41B active) and native multimodal support for text, image, and audio input with text output. It supports up to a 1M-token context window and controllable reasoning effort, letting developers trade off cost and latency. Trained on 45 trillion tokens, it targets broad, balanced performance across reasoning, agentic coding, tool use, and multimodal tasks.
About Thinking Machines Lab
View Company Profile
Last updated: July 16, 2026
