Thinking Machines Unveils Inkling Small Model

Thinking Machines, the AI startup led by former OpenAI CTO Mira Murati, has released a new open-source model called Inkling-Small. The model is designed to deliver performance close to its predecessor, Inkling, but at roughly one-quarter the size. The release comes just two weeks after the original Inkling, signaling a rapid iteration cycle from the company. Despite its smaller footprint, Inkling-Small actually outperforms its larger sibling in several benchmarks. This is a notable achievement, as it demonstrates that efficiency does not necessarily come at the cost of capability. The model's compact design makes it easier to deploy in resource-constrained environments, such as mobile devices, edge servers, or applications with limited memory and compute budgets. The open-source nature of Inkling-Small is a key part of Thinking Machines' strategy. By making the model freely available, the company aims to attract developers and researchers who want to experiment with efficient AI without the overhead of proprietary systems. This move also positions Thinking Machines as a player in the open-source AI community, competing with other releases from organizations like Meta and Mistral. Inkling-Small is built on the same architectural principles as the original Inkling but with optimizations that reduce parameter count and computational requirements. The company says the model maintains strong reasoning and language understanding capabilities, making it suitable for a wide range of tasks, from chatbots to code generation. For developers, the release of Inkling-Small offers a practical option for deploying AI in settings where larger models are impractical. It also provides a benchmark for what can be achieved with careful model compression and distillation techniques. Thinking Machines plans to continue iterating on the Inkling family, with a focus on balancing performance, size, and accessibility.