Open Models Thrive: Laguna S2.1, Inkling, Kimi K3 Emerge

New open models like Laguna S2.1, Inkling, and Kimi K3 highlight the growing power of open models. Despite predictions of consolidation, more companies are now training and openly releasing strong models. Here's the thing: The capacity to train powerful models is rapidly increasing. Many thought consolidation was inevitable due to rising training costs. Yet, we see more companies, investing hundreds of millions to billions, releasing models openly. Demand for tokens is very high and likely to grow. Companies once expected to consolidate are finding value in building "token machines." For example, Thinking Machines, which launched in February 2025, now has an open model finetuning service generating hundreds of millions in revenue annually. Their open-weight models are leading in the U.S.A. What's interesting is the sustained pace from Chinese labs, with new entrants like Xiaomi gaining influence. Instead of consolidation, continued adoption seems a safer bet, with open models playing a key role. Inkling by Thinking Machines is a 975B-A41B multimodal model that takes text, images, and audio as input and produces text. It's designed as a strong base for fine-tuning. Tencent’s Hy3, a 295B-A21B model, shows improvement and now uses the Apache 2 license. The bottom line is that the landscape of AI model development is rapidly evolving, with open models taking an increasingly central role.