Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, has introduced its first model, Inkling. The news matters not only because of who founded the company, but because Inkling is a multimodal model built on a mixture-of-experts (MoE) architecture with 975 billion parameters. For anyone tracking frontier AI models, this signals a strong new player entering the market, one capable of competing with the leaders.
What happened
Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, has presented its first model, named Inkling. It is a multimodal model built on a mixture-of-experts (MoE) architecture and counting 975 billion parameters. Only a fraction of those parameters is active per token, the hallmark of MoE architectures, which is what makes high efficiency possible at such a large total parameter count. The company spent close to a year and a half getting to this release.
Why it matters
The arrival of Inkling from Mira Murati's startup is a significant event for the AI industry. Multimodal MoE models are one of the most promising directions in large language model development, offering a balance between performance and compute cost. Given Murati's track record at OpenAI, it is reasonable to expect Inkling to carry frontier capabilities and to become a serious competitor to the existing flagship models. It also confirms the trend toward decentralised frontier AI development, where new players with a strong team can climb to leading positions quickly.
What it means in practice
- A powerful new tool: as a multimodal MoE model, Inkling will likely offer distinctive capabilities across different data types (text, image, audio) and handle complex tasks that require integrating information from several modalities.
- Cost optimisation: the MoE architecture delivers high performance at lower inference compute cost than dense models of comparable size, which could make Inkling more accessible to developers and companies.
- Sharper competition: Inkling's arrival intensifies competition among foundation model builders, which ultimately drives better quality, lower prices and new innovative solutions.
- Research opportunities: for researchers and developers working with frontier AI, Inkling becomes a new platform for experiments and products, especially in areas where multimodality is central.
- Watch the benchmarks: it is worth following public benchmarks and comparisons of Inkling against other top models closely to understand its real advantages and best use cases.