Breaking

Thinking Machines Lab Releases Inkling, an Open-Weight Model Built for Customization

July 15, 2026 at 14:12 EDT

Foundation Models
Open Source
Sales & Business

Thinking Machines Lab, the startup led by former OpenAI CTO Mira Murati, has released Inkling, its first foundation model, betting that enterprises want an open-weight system they can tailor to their own data rather than renting access to closed frontier models. The model launched on July 15, 2026, with its full weights published on Hugging Face and immediate fine-tuning available through the company's Tinker platform.

July 15, 2026 · Thinking Machines Lab

Inkling: Mira Murati's Lab Bets on Open Weights

The startup's first foundation model ships with full weights on day one — a wager that enterprises want a system they can host, tune, and control rather than rent from closed frontier labs.

975B

total parameters (mixture-of-experts)

41B

active parameters per query

token context window

45T

tokens pretrained across modalities

Capacity vs. cost: only a fraction runs per query

The MoE design keeps a very large model but activates just ~4% of it on each request.

975B

Total
capacity

41B

Active
per query

The self-improvement loop

Inkling generates its own data

→

Fine-tunes on that data

→

Stronger model feeds Tinker ecosystem

Early praise

Strong, customization-friendly open-weight base
Well suited to local tuning
Responses feel emotionally mature; easy to steer

Open questions

Common-sense blind spots remain
Heavy token consumption
Everyday judgment still uneven

The bet

The next competitive front isn't just raw capability — it's who owns and controls the model.

IP protection
Keep proprietary data in-house

Cost control
Host instead of renting access

On-premises
Deploy and modify freely

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.