Digital Event Horizon
Together AI has announced the release of Inkling, a groundbreaking multimodal mixture-of-experts model designed for advanced reasoning. With its ability to accept multiple input modalities and provide strong performance across a wide range of tasks, Inkling is poised to revolutionize applications in fields such as visual question answering, document analysis, and audio comprehension. Learn more about this exciting development in the field of artificial intelligence.
Inkling, a cutting-edge model, has been released on Together AI's platform with advanced reasoning capabilities. The model combines benefits from different attention mechanisms for strong reasoning and multimodal capabilities. Inkling can accept multiple input modalities, including text, image, and audio, enabling a wide range of applications. The model has demonstrated strong performance across various benchmarks and tasks. Together AI's production inference platform provides a scalable environment for running Inkling.
Together AI, a cutting-edge artificial intelligence platform, has recently announced its collaboration with Thinking Machines Lab to bring on board the latest model, Inkling. This innovative multimodal mixture-of-experts model is designed specifically for advanced reasoning and has been made available on Together AI's production inference platform. The release of Inkling marks a significant milestone in the development of artificial intelligence, as it introduces several architectural innovations that enable strong reasoning and multimodal capabilities while maintaining efficient model execution.
Inkling is a decoder-only mixture-of-experts model with 975B total parameters, 40B active parameters per token, and a context window of 1M tokens. Its design combines the benefits of different attention mechanisms, including query-conditioned relative attention, short causal convolutions, and a shared expert sink. These components work together to provide the model with a flexible mechanism for representing token order and nearby context.
One of the key features of Inkling is its ability to accept multiple input modalities, including text, image, and audio. Lightweight embedding towers transform these inputs into embeddings that are inserted directly into the model's input sequence, allowing the language model to reason jointly over different types of information. This unified design enables a wide range of applications, such as visual question answering, document analysis, audio comprehension, multimodal agents, and workflows that combine multiple input types within the same conversation.
Inkling has been evaluated on a variety of benchmarks, including scientific reasoning, mathematics, coding, agentic, vision, and audio. Its results demonstrate strong performance across these tasks, showcasing its versatility and broad task versatility. The model is also post-trained on forecasting and calibrated prediction tasks, expanding its usefulness beyond conventional question answering to applications where representing uncertainty and producing well-calibrated estimates are important.
Together AI's production inference platform provides a scalable and reliable environment for running Inkling. With day 0 access, developers can build text, image, and audio applications using a unified model, scale from initial experimentation to dedicated production capacity, and control the reasoning effort through the API directly. This enables cost and speed tuning at the request level, not the infrastructure level.
The release of Inkling on Together AI's platform is an exciting development in the field of artificial intelligence. With its advanced reasoning capabilities and multimodal design, this model has the potential to revolutionize a wide range of applications and industries. As part of its commitment to supporting innovation and collaboration, Together AI continues to push the boundaries of what is possible with artificial intelligence.
Related Information:
https://www.digitaleventhorizon.com/articles/Together-AI-Unveils-Inkling-A-Groundbreaking-Multimodal-Mixture-of-Experts-Model-for-Advanced-Reasoning-deh.shtml
https://www.together.ai/blog/together-ai-brings-thinking-machines-labs-new-model-inkling-on-day-0
Published: Wed Jul 15 13:15:43 2026 by llama3.2 3B Q4_K_M
