Home / Technology / AI Runs Offline: Inkling 2 Mini on Your Laptop
AI Runs Offline: Inkling 2 Mini on Your Laptop
22 Nov
Summary
- Inkling 2 Mini runs offline on consumer hardware.
- New architecture offers deep reasoning without large size.
- Apache 2.0 license encourages broad enterprise adoption.

Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, has launched Inkling 2 Mini, an open-weights multimodal AI model capable of running offline on standard consumer hardware. This U.S.-built model, licensed under Apache 2.0, utilizes a unique 'looped transformer' architecture to achieve strong reasoning performance without requiring vast parameter counts or cloud infrastructure.
Inkling 2 Mini boasts a 26.4 billion parameter count and a ~20GB memory footprint, rivaling larger systems on benchmarks for software engineering and tool use. Its innovative architecture processes data through middle layers multiple times per token, effectively increasing depth without increasing storage. This approach allows for efficient offline operation, as demonstrated by its ability to power applications like a flight simulator on a MacBook Pro.
The model's performance is notable on agentic coding and tool-use evaluations, outperforming other open-weights models in its class. While it may trail denser models in factual recall, its local processing power and cost-effectiveness present a compelling alternative for developers and enterprises seeking private, on-premises AI solutions. The company previously released Inkling and Inkling-Small, with Inkling 2 Mini representing a significant step towards practical local AI deployment.