Thinking Machines, the startup founded by former OpenAI CTO Mira Murati, has released Inkling, a 975B parameter model that processes text, images, and audio. The model is currently the most powerful U.S. open-weights model, outperforming competitors like Kimi K2.6 and DeepSeek v4 Flash max on agentic tasks.

Inkling is a multimodal open-weights model developed by Thinking Machines Lab. It is designed to handle a wide range of tasks, including text, image, and audio processing. The model's release has been met with interest from the AI community, though it faces competition from other large language models.

According to the analysis platform Artificial Analysis, Inkling outperforms U.S. models like Kimi K2.6 and DeepSeek v4 Flash max on agentic tasks. However, it trails behind Chinese models in certain benchmarks. The model has 975 billion parameters, making it one of the largest models available in the open-weights category.

Despite its strong performance in some areas, Inkling has notable weaknesses, including a 63% hallucination rate. This high rate of factual inaccuracies could limit its practical applications. Additionally, the model comes at a higher cost compared to comparable Chinese models, which may affect its adoption in cost-sensitive environments.

The release of Inkling highlights the ongoing competition in the AI space, with U.S. models striving to match the capabilities of their Chinese counterparts. While Inkling represents a significant advancement in open-weights models, its higher cost and hallucination rate may influence its market reception and long-term viability.