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

Thinking Machines Lab, the innovative AI startup led by former OpenAI CTO Mira Murati, has unveiled its first proprietary AI model named Inkling. Released on Wednesday, this groundbreaking model sets itself apart from leading models offered by OpenAI, Anthropic, and Google by being open-weight. This feature allows external developers and companies to download and modify Inkling to suit their needs. Inkling is a mixture-of-experts system boasting a staggering 975 billion parameters; however, it utilizes only a portion—around 41 billion—for specific tasks. This efficient design not only accelerates performance but also reduces operational costs. Trained on a diverse dataset of 45 trillion tokens, encompassing text, images, audio, and video, Inkling can reason across all three modalities. This release marks a significant milestone for the company, following a year and a half of behind-the-scenes development of AI infrastructure. Previously, the company provided a glimpse of its capabilities in a May research preview showcasing "interaction models"—AI that engages in conversation more fluidly than traditional chatbots. The core premise behind Thinking Machines is that customizable AI solutions will outperform the one-size-fits-all models currently prevalent in the market. Inkling is designed to provide calibrated responses, enabling users to adjust the "thinking effort" according to their requirements, opting for speed when necessary. In one benchmark, the company claims that Inkling achieves comparable coding performance while consuming a third of the tokens used by Nvidia’s Nemotron 3 Ultra. While Thinking Machines does not position Inkling as the leading model available, it emphasizes well-rounded performance. The company is targeting enterprises, marketing Inkling as a starting point for organizations to refine and adapt through Tinker, its model-customization platform. This approach diverges from the strategies of OpenAI, Anthropic, and Google, which primarily focus on creating general-purpose chatbots. A blog post released prior to Inkling’s launch argued against the limitations of centrally-trained AI models. The company believes that AI tailored by organizations can provide superior performance, as these models incorporate specific expertise. This perspective is gaining traction within the industry, as Microsoft CEO Satya Nadella recently highlighted the dual costs associated with proprietary AI models in a blog post. Recent collaborations, such as the one with Bridgewater Associates, have demonstrated the potential of open-source models. Researchers found that by further training an existing open-source model with Bridgewater's financial expertise, they achieved an impressive 84.7% score on financial reasoning tests, outperforming top proprietary models while being significantly more cost-effective. Thinking Machines has also noted its rapid development pace, claiming to bring its technology to market in just nine months—much quicker than its competitors. Questions remain regarding whether Inkling was trained using outputs from rival models, a process known as distillation. According to the company, Inkling was primarily pretrained from scratch but did incorporate data from other open-weight models to enhance its training process. On the financial front, the company has been cautious. A strategic partnership with Nvidia aims to leverage substantial computing resources, yet specific revenue strategies remain undisclosed. Additionally, there is speculation about whether Thinking Machines will ever reach the financial scale of OpenAI or Anthropic, or if its efficiency-focused model will allow it to operate differently. As it stands, Thinking Machines employs around 200 people, recovering from earlier staffing challenges. The company prioritizes stability and continuity over high-profile talent acquisitions, which aligns with its overall approach to fostering a collaborative work environment. This strategy may prove advantageous as the company continues to navigate the competitive landscape of AI development.