Using Inkling

Inkling is Thinking Machines Lab's model tailored for Tinker. It is a general-purpose model that can code, reason, and call tools. Model effort can be controlled by a float. Image and audio inputs are also supported.

Use tml-renderers to construct all of these inputs.

Messages can be OpenAI-compatible Cookbook dictionaries or typed tml_renderers.chat objects. Use the format that matches your application.

Get started

Install the Cookbook with the inkling extra, which pulls in the tml_renderers package that Inkling's renderer and scripts depend on:

uv pip install 'tinker-cookbook[inkling]'

Set the model name in your existing Cookbook workflow:

model_name = "thinkingmachines/Inkling"

We recommend sweeping learning rate and thinking effort to adapt Inkling to your use case. Use sample_reasoning.py to compare standard effort values.

The Cookbook selects Inkling's recommended renderer automatically. Each guide links to its corresponding runnable example.

Learn more

• tml-renderers — Message formats, tools, rendering, and training.
• Thinking effort — Effort controls, recommended sweep values, and scaling results.
• Audio and Images — Multimodal input requirements.
• Rendering tutorial — How renderers bridge messages and model tokens.
• Tinker sampling quickstart — The lower-level sampling flow.