What's Happening?
Thinking Machines, an AI startup founded by OpenAI’s former chief technology officer Mira Murati, has launched Inkling 2 Mini, an open-weights artificial intelligence model designed to run efficiently on personal laptops. This new model, which is small
enough for a MacBook, achieves performance comparable to larger rivals like Alibaba’s Qwen3.8 27B, according to independent rankings by Artificial Analysis. Inkling 2 Mini utilizes a 'looped transformer' architecture, employing 24 layers but running the middle 12 layers three times for each token. This design allows it to operate with 26.4 billion stored parameters, with only 3.4 billion active per token (or 5.4 billion when loops are counted), making it highly efficient. The model requires approximately 20 gigabytes of memory and is compatible with Macs having at least 32 gigabytes of RAM, as well as Nvidia’s DGX Spark mini supercomputer. Its weights are available on Hugging Face under the Apache 2.0 license, and its API costs $0.15 per million input tokens and $0.50 per million output tokens.
Why It's Important?
The release of Inkling 2 Mini marks a significant advancement in making powerful AI models more accessible and efficient for individual users and smaller enterprises in the U.S. Its ability to run locally on a laptop means that data processing occurs on-device, enhancing privacy and security for sensitive information, which is a critical concern for businesses and developers. The model's efficiency, being six times cheaper and four times faster per task than comparable models like Qwen, could significantly reduce operational costs and accelerate development cycles for AI applications. This efficiency, combined with its strong performance in coding and agentic tasks, positions Inkling 2 Mini as a valuable tool for software development, data analysis, and other AI-driven applications, potentially fostering innovation and competition within the U.S. tech industry by lowering the barrier to entry for advanced AI utilization.
What's Next?
The availability of Inkling 2 Mini's weights under the permissive Apache 2.0 license means that developers and companies can freely use and fine-tune the model for commercial purposes. This open-source approach is likely to encourage widespread adoption and customization, potentially leading to a proliferation of new AI applications and services. However, Thinking Machines acknowledges certain limitations, such as the model's verbosity and slower start-up time compared to some rivals, as well as a higher non-hallucination rate. Future developments will likely focus on addressing these weaknesses, improving knowledge retention, and optimizing performance for multi-turn chat and complex instructions. The looped architecture also raises questions about monitoring internal computations, which Thinking Machines is actively researching, suggesting ongoing efforts to balance efficiency with transparency and safety in AI development.
Beyond the Headlines
Inkling 2 Mini's looped transformer architecture represents a novel approach to AI model design, prioritizing computational efficiency without sacrificing performance. This innovation could influence future AI research and development, pushing the industry towards more resource-optimized models that can run on less powerful hardware. The emphasis on local execution for privacy and security could also set a new standard for AI deployment, particularly in sectors dealing with highly sensitive data, such as healthcare, finance, and government. While the model's current limitations in knowledge and non-hallucination rate are noted, the underlying architectural advancements suggest a pathway towards more intelligent and trustworthy AI systems that are both powerful and practical for everyday use. This development could democratize access to advanced AI capabilities, moving beyond reliance on large, cloud-based models and empowering a broader range of innovators.













