NewsBite analysis · AI, Startups
VB
Thinking Machines debuts Inkling Small open source AI model nearing performance of predecessor at about 1/4 size
Thinking Machines unveiled Inkling-Small, an open-source AI model aimed at enterprises. This 276-billion-parameter model closely matches the performance of the larger 975-billion-parameter Inkling model while drastically reducing computational demands. Though slightly lagging in factual understanding, Inkling-Small excels in various benchmarks, offering an attractive balance between performance and efficiency, particularly in multimodal applications. Enterprises can leverage the model without the extensive compute resources needed for its larger counterpart.
ExperimentConfidence: HighShort term horizon
Published 31 Jul 2026, 00:25 UTC7 min read
Visit original article ↗Lead image — VB
Summary
TL;DR
- Inkling-Small is a 276-billion-parameter AI model, significantly smaller than Inkling's 975 billion.
- It maintains comparable performance levels across multiple benchmarks.
- The model is open-source under the Apache 2.0 license, allowing flexibility for enterprise use.
- Inkling-Small uses 12 billion active parameters per token for efficient computation.
- Favorable for coding assistants and multimodal applications, despite limitations in factual recall.
- Deployment is more manageable, targeting enterprise GPU servers versus consumer-grade hardware.
Interesting facts
- Inkling-Small scored higher on several benchmarks compared to its larger counterpart.