NewsBite analysis · AI, Programming, Startups
The Decoder
Ex-OpenAI CTO Murati's Thinking Machines drops Inkling, a 975B parameter model that leads US labs but trails China
Thinking Machines Lab has unveiled Inkling, a 975 billion parameter open-weights model designed for efficiency and flexibility in AI tasks. It holds the title of the leading open-weights model from a U.S. lab, outscoring competitors like Nemotron 3 Ultra. Despite its capabilities, Inkling does not match the performance of leading Chinese models, particularly in factual accuracy, where it scored just +2 on the Omniscience benchmark. This raises concerns about its usability in applications requiring precise outputs.
MonitorConfidence: MediumShort term horizonData governance risk: High
Published 16 Jul 2026, 10:10 UTC3 min read
Visit original article ↗Lead image — The Decoder
Summary
TL;DR
- Inkling features 975 billion parameters, with 41 billion active at once.
- It natively processes text, images, and audio, supporting up to one million tokens.
- The model was pre-trained on 45 trillion tokens, utilizing a mix of public and synthetic data.
- It ranks highest among U.S.-made models, but trails behind Chinese models in overall performance.
- Inkling-Small, a 276 billion parameter variant, shows promise with competitive results.
Interesting facts
- Inkling's architecture allows for a flexible balance between performance and cost.
- It uses fewer output tokens than comparable open-weights models for similar tasks.