Thinking Machines

Mira Murati's Thinking Machines Lab released Inkling-Small, a 276-billion-parameter mixture-of-experts model with 12 billion active parameters that approaches the performance of its much larger Inkling sibling. The open-weights multimodal model is available on Hugging Face and can be fine-tuned through the company's Tinker platform.

Thinking Machines Lab shipped Inkling, its first open-weight model, a 975-billion-parameter mixture-of-experts system trained on 45 trillion multimodal tokens. The launch is the startup's first public proof point after 18 months of building, and it monetizes through Tinker, its fine-tuning platform, betting that customized models beat generic frontier AI.

The Next Web reports that Thinking Machines Lab openly concedes Inkling is not the strongest model on the market, positioning it instead as a foundation for organizations to fine-tune on their own data. A partnership with hedge fund Bridgewater, where a customized version scored 84.7% on financial reasoning benchmarks, illustrates the strategy.