NewsBite analysis · AI, Business
Silicon Republic
Thinking Machines Lab unveils new Inkling version at 25pc size
Thinking Machines Lab has unveiled Inkling-Small, a new AI model that promises comparable performance to the original Inkling but at 25% of its size. Backed by Nvidia, the model features 276 billion total parameters and offers enhanced speed and efficiency, achieving high scores on various benchmarking tests. Designed for practical applications in text, image, and audio processing, it aims to improve usability in complex document handling, demonstrating the company's commitment to developing adaptable AI systems.
ExperimentConfidence: HighShort term horizon
Published 31 Jul 2026, 14:55 UTC2 min read
Visit original article ↗Lead image — Silicon Republic
Summary
TL;DR
- Inkling-Small offers similar performance to Inkling at a fraction of its size.
- The model has 276 billion parameters and runs efficiently with 12 billion active.
- It achieves a benchmark score of 40% on the Intelligence Index.
- Faster than the average model, it processes 93 tokens per second.
- Designed for real-world applications, especially in document usability.
Interesting facts
- Inkling-Small was developed following a significant investment partnership with Nvidia.
- The model processes text, image, and audio inputs, outputting text.
- It demonstrates noteworthy scalability in performance while reducing compute requirements.