Thinking Machines amps up its bet against one-size-fits-all AI with its first open model, Inkling

Key Points:

• Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, launched Inkling, its first open-weight AI model with 975 billion parameters, allowing external developers to download and modify it directly.
• Inkling uses a mixture-of-experts architecture, drawing on about 41 billion parameters per task, and was trained on 45 trillion tokens across text, image, audio, and video, though its outputs are currently limited to text-based formats.
• The model emphasizes calibrated answers with uncertainty flags and adjustable "thinking effort," targeting enterprises that want customizable AI rather than one-size-fits-all solutions from competitors like OpenAI, Anthropic, and Google.
• Thinking Machines positions Inkling as a starting point for organizations to fine-tune via its Tinker platform, placing responsibility for safe customization on users and focusing revenue on model customization and hosting services rather than metered access.
• The company highlights its rapid development pace—bringing Inkling to market in about nine months—and argues that adaptable AI models can outperform closed, centrally trained ones, citing a successful collaboration with Bridgewater Associates as evidence.