Thinking Machines' Inkling: Impressive Base Model or Not Ready Yet?

Description

I tested Thinking Machines Lab's Inkling live across branding, website creation, code reasoning, general knowledge, and an enterprise compliance scenario.

Inkling is an open-weights mixture-of-experts model with 975B total parameters and 41B active. But scale is not the real question. The useful question is whether the model can follow instructions, explain its reasoning, produce practical work, and identify risks that matter in real applications.

In this episode, you will see:

- A general-knowledge answer checked against visible sources

- A coffee brand and interactive website concept created without image references

- The generated website's strongest ideas and its readability failure

- Inkling rejecting a hackish coding shortcut in its own reasoning

- A realistic SaaS compliance-assistant test

- Why auditability and context quality matter for production AI

- My verdict on using Inkling for product work and agent stress testing

Learn more about Inkling:

- Inkling overview: https://thinkingmachines.ai/inkling/

- Official announcement: https://thinkingmachines.ai/news/introducing-inkling/

- Model card: https://thinkingmachines.ai/model-card/inkling/