Friday, July 17, 2026

Today’s Edition

AI Intel Report

MARKETS

Frontier Models

Thinking Machines Lab Releases Inkling 975B Open-Weights Multimodal MoE

Ex-OpenAI CTO Mira Murati's lab introduces a customizable base model that prioritizes multimodal breadth and fine-tuning flexibility over benchmark dominance in frontier AI.

3 MIN READ
Inside a spacious open-plan technology research laboratory with floor-to-ceiling windows overlooking an urban skyline the scene shows multiple rows of black server racks filled with visible GPU cards and tangled multicolored cabling running along the floor and ceiling trays three anonymous technicians wearing white lab coats stand with backs to the viewer one technician kneels adjusting a cable bundle on the lowest rack another stands reaching into an open server chassis revealing internal cooling fans and circuit boards a third technician points toward a central workstation table holding three large flat panel monitors displaying grids of abstract colorful image patches waveform patterns and layered data visualizations without any readable characters or symbols scattered across the table are disassembled hardware modules including heatsinks power supplies and empty model training cartridges a fourth technician sits at a side desk typing on a keyboard connected to a laptop that shows split screen views of photographic scenes and sensor readouts the floor features anti-static mats and rolling equipment carts loaded with spare networking switches and fiber optic spools overhead industrial lighting fixtures cast even illumination across the entire workspace while background elements include additional unoccupied desks with closed laptops charging cables and reference binders the overall environment conveys a functional high density computing facility dedicated to large scale model development with emphasis on hardware accessibility and collaborative physical interaction among the figures all elements remain strictly within the bounds of a single continuous live action photograph capturing the tangible infrastructure supporting multimodal mixture of experts model releases and open weight distribution practices
Illustration: AI Intel Report

Inkling is a decoder-only multimodal Mixture-of-Experts transformer with 975B total parameters and 41B active parameters released by Thinking Machines Lab.

Ex-OpenAI CTO Mira Murati founded Thinking Machines Lab to develop open-weights models that emphasize customization and efficiency.

What background led to the development of Inkling?

The lab focuses on alternatives to closed frontier models by releasing base models suited for fine-tuning across diverse applications.

This strategy addresses developer needs for adaptable systems rather than fixed closed offerings.

What is the technical architecture of Inkling?

Inkling uses a 66-layer decoder-only transformer with a sparse Mixture-of-Experts feed-forward backbone.

The design includes 256 routed experts and 2 shared experts, activating 6 routed experts per token.

Images receive encoding through a hierarchical patch encoder while audio uses discrete token encoding.

Relative attention replaces RoPE and hybrid sliding-window plus global attention layers manage long sequences.

Controllable thinking effort lets users trade performance against token usage during inference.

How was Inkling pretrained and what context does it support?

Pretraining drew from 45 trillion tokens spanning text, images, audio, and video.

The context window extends to 1 million tokens for handling extended input sequences.

Native multimodal reasoning processes text, images, and audio inputs directly to produce text outputs.

What market implications arise from the Inkling release?

Open weights under Apache 2.0 on Hugging Face enable broad access for customization by researchers and enterprises.

Integration with the Tinker platform supports targeted fine-tuning workflows.

The approach offers developers an open alternative to closed models that restrict adaptation.

What expert reactions address the model's positioning?

The company positions Inkling as a practical base rather than a benchmark leader.

Inkling is not the strongest overall model available today, open or closed. Instead, a combination of qualities makes it a good open-weights base for customization: multimodal capabilities, efficient thinking, and availability on Tinker for fine-tuning.Thinking Machines Lab

What comes next for the Inkling model family?

Inkling serves as the initial release in a planned family of models.

A preview of the smaller Inkling-Small variant at 276B total and 12B active parameters has been shared.

Inkling model family specifications
VariantTotal ParametersActive ParametersContext WindowLicense
Inkling975B41B1M tokensApache 2.0
Inkling-Small276B12B1M tokensApache 2.0
  1. Access weights via the Hugging Face repository under Apache 2.0.
  2. Review the model card for architecture and encoding details.
  3. Deploy on Tinker for fine-tuning experiments.
  4. Adjust controllable thinking effort for efficiency tuning.
  5. Test native multimodal inputs including images and audio.

Frequently asked

What pretraining scale supports Inkling capabilities?

Inkling was pretrained on 45 trillion tokens of text, images, audio and video.

Where is the Inkling model available for use?

The model is released with open weights under Apache 2.0 on Hugging Face and supports fine-tuning on the Tinker platform.

How does Inkling handle multiple input types?

Inkling reasons natively over text, images, and audio inputs through dedicated encoders while outputting text.