Thinking Machines, founded by OpenAI's former CTO Mira Murati, has launched Inkling, a 975-billion-parameter general-purpose model with a 1-million-token context window. The model is designed to be efficient, flexible, and cost-effective for enterprise users.
TL;DR
- Thinking Machines introduces Inkling, a 975B-parameter model with a 1M-token context window.
- Inkling uses controllable thinking to reduce token usage by 66% compared to competitors like Nemotron 3 Ultra.
- The model supports native multimodality, processing images, videos, and audio together for better efficiency.
What happened
Thinking Machines, an AI startup founded by former OpenAI CTO Mira Murati, introduced Inkling on Wednesday. Inkling is a general-purpose model designed to be efficient, flexible, and cost-effective for enterprise users. The model is an open-weight model that supports controllable thinking, allowing users to adjust the model's reasoning to control costs.
Inkling is a Mixture-of-Experts model with 975 billion total parameters. It supports a context window of up to 1 million tokens and was pretrained on 45 trillion tokens of text, images, audio, and video. According to the company, Inkling uses one-third as many tokens as models such as Nemotron 3 Ultra by compressing its thought process.
The model also features native multimodality, processing images, videos, and audio together as a continuous whole. This approach is designed to make the model more efficient and reduce hallucinations.
Why it matters
Inkling's efficiency and cost-effectiveness make it an attractive option for enterprises looking to implement AI models. The model's ability to control costs by adjusting reasoning could be particularly appealing to companies with limited resources.
However, Thinking Machines will need to differentiate itself from established players like OpenAI, Anthropic, and open-source model providers like Nvidia. Analysts suggest that the company may need to partner with specific hardware providers or focus on specific use cases to stand out in the competitive AI market.
For developers and startups, Inkling's open-weight model and support for controllable thinking could provide a valuable tool for building AI applications. The model's efficiency and cost-effectiveness could make it an attractive option for companies looking to implement AI models without breaking the bank.
Key facts
- Inkling is a 975-billion-parameter Mixture-of-Experts model.
- The model supports a context window of up to 1 million tokens.
- Inkling was pretrained on 45 trillion tokens of text, images, audio, and video.
- According to the company, Inkling uses one-third as many tokens as models such as Nemotron 3 Ultra.
- The model features native multimodality, processing images, videos, and audio together.
- Thinking Machines was founded by former OpenAI CTO Mira Murati.
- Analysts suggest that Thinking Machines may need to partner with specific hardware providers or focus on specific use cases to differentiate itself from competitors.
- Inkling is designed to be efficient, flexible, and cost-effective for enterprise users.
Context
The launch of Inkling comes as the AI market continues to grow rapidly. Enterprises are increasingly looking to implement AI models to improve efficiency and reduce costs. However, the high cost of training and running large AI models has been a barrier to adoption for many companies.
Inkling's efficiency and cost-effectiveness could make it an attractive option for enterprises looking to implement AI models. The model's ability to control costs by adjusting reasoning could be particularly appealing to companies with limited resources.
However, Thinking Machines will face stiff competition from established players like OpenAI, Anthropic, and open-source model providers like Nvidia. The company will need to differentiate itself through partnerships, specific use cases, or other innovations to stand out in the crowded AI market.