Thinking Machines Lab has released its first open-weights model, Inkling, with 975B parameters and multimodal input capabilities [1]. The model is designed to accept text, image, and audio inputs and generate text outputs, making it suitable for various applications such as agentic and tool-use systems, coding assistants, chatbots, and retrieval-augmented generation systems [2]. Inkling is available on Tinker, Hugging Face, and other platforms, allowing researchers and startups to download and modify it [3]. The model's capabilities include generalist reasoning, agentic coding and tool use, multimodality, epistemics, and safety [4].
Inkling has been trained on a large dataset of 45 trillion tokens and has undergone large-scale asynchronous RL to shape its behavior and improve its performance [5]. The model's performance has been evaluated on various benchmarks, including HLE, AIME, GPQA, and others, and has shown competitive results [6]. Inkling-Small, a smaller version of the model with 12B active parameters, has also been released and matches or exceeds Inkling's performance on many benchmarks [7]. The release of Inkling marks a significant milestone for Thinking Machines Lab, which aims to establish itself as a major player in the AI industry [8].
