Thinking Machines Launches Inkling, a 975-Billion Parameter Open-Weight Multimodal Model
Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, has released its first open-weight AI model, Inkling, designed for enterprise customization. The 975-billion parameter mixture-of-experts model natively supports multimodal reasoning across text, image, audio, and video while optimizing for cost and efficiency. This release marks a strategic shift toward decentralized, fine-tuned corporate AI models rather than centralized proprietary services.
Skynet Chance (+0.01%): Releasing highly capable, open-weight models increases the risk of unaligned or malicious modification since safety guardrails can be stripped away by end-users. However, because Inkling is not a frontier-shattering model, the overall immediate increase in existential risk remains relatively minor.
Skynet Date (-1 days): Decentralized open-weight development accelerates the deployment of capable AI across various industries outside of centralized oversight. This rapid, uncontrolled distribution potentially brings forward the timeline for unpredictable systemic AI failures or loss-of-control scenarios.
AGI Progress (+0.02%): The model's native multimodal reasoning, mixture-of-experts efficiency, and fast training timeline demonstrate significant practical progress in building and scaling complex AI architectures. Its open-weight nature also allows the global research community to build upon these advancements, further driving collective AGI research.
AGI Date (-1 days): The incredibly fast nine-month development cycle from a new startup, combined with native multimodal capabilities, indicates that the barrier to entry for training massive models is dropping. This rapid development pace, fueled by advanced computing partnerships, accelerates the overall projected timeline to achieving AGI.