Person sketching rapidly on a small tablet, capturing a complex scene with speed and precision

Inkling 2 Mini: Local AI Power

Inkling 2 Mini is a new model from Thinking Machines Lab. It is designed to be fast and powerful while being small enough to run on a laptop. The model uses a looped transformer architecture to improve reasoning and efficiency.

Thinking Machines Lab released Inkling 2 Mini, a small, powerful model designed to run on laptops and other local hardware, excelling in agentic coding tasks.

A small, powerful sports car speeds through a mountain road, outperforming larger vehicles with ease

TL;DR

Inkling 2 Mini is a new model from Thinking Machines Lab. It is designed to be fast and powerful while being small enough to run on a laptop. The model uses a looped transformer architecture to improve reasoning and efficiency.

Key facts

  • Maker: Thinking Machines Lab
  • What's new: Looped transformer architecture for improved reasoning and efficiency
  • Access: Available on Tinker and Hugging Face
  • License: Open weights
  • Size: 26.4B total parameters, ~20GB memory footprint
  • Context: Runs offline on consumer laptops and desktop AI computers
  • Benchmark: DeepSWE v1.1 (48.2%), Vals Index v2.2 (46.5%)
Artisan assembling a complex, miniature clockwork mechanism by hand, showcasing intricate design and craftsmanship

What is it?

Inkling 2 Mini is a fast and powerful AI model from Thinking Machines Lab. It is designed to be small enough to run on local devices like laptops. The model uses a looped transformer architecture, which means its middle layers run multiple times for each token.

How does it work?

The looped transformer design allows Inkling 2 Mini to apply more reasoning with fewer tokens. This helps it reach correct answers more efficiently, especially for tasks like agentic coding. The model's weights are deployed in NVFP4 and MXFP8 precision, contributing to its small memory footprint.

Why does it matter?

Inkling 2 Mini offers high performance for its size, leading open-weights models under 100B parameters on key benchmarks. Its ability to run on local hardware makes it suitable for persistent agents and offline use. The looped architecture also improves monitorability, allowing better oversight of the model's reasoning.

Who is it for?

Developers, researchers, and users who need a powerful AI model that can run efficiently on local hardware for tasks like agentic coding and knowledge work.

Try it

Available on Tinker and Hugging Face

Frequently asked questions

What is Inkling 2 Mini?

Inkling 2 Mini is a fast, powerful AI model released by Thinking Machines Lab. It is designed to be small enough to run on consumer laptops and desktop AI computers. The model uses a looped transformer architecture to enhance its reasoning capabilities.

How does Inkling 2 Mini improve reasoning?

Inkling 2 Mini improves reasoning through its looped transformer architecture. The middle layers of the model run three times for every token, allowing it to apply more computation internally. This design helps the model reach accurate answers with fewer tokens, especially for tasks requiring high reasoning quality.

Can Inkling 2 Mini run offline?

Yes, Inkling 2 Mini is designed to run on local devices such as consumer laptops and desktop AI computers. It has a total memory footprint of approximately 20GB, making it suitable for offline operation. It was demonstrated running offline on a MacBook Pro.

What are the strengths of Inkling 2 Mini?

Inkling 2 Mini excels at long-horizon agentic work and agentic coding benchmarks like DeepSWE v1.1 and SWE-bench Pro v2. It also performs well on knowledge work benchmarks that involve search and tool use. Its efficiency and performance make it a strong choice for running multiple persistent agents on local hardware.

Sources

  1. Thinkingmachines

Follow news like this automatically

RAI AI watches news, X, Telegram and Reddit for the topics you choose and sends a short AI summary to the app, email or Telegram. Try RAI AI free for 7 days