Ex-OpenAI CTO Murati’s Thinking Machines drops Inkling, a 975B parameter model that leads US labs but trail…
What happened
Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, launched Inkling, a new open-weights AI model with 975 billion parameters. Inkling is a multimodal model, capable of processing different types of input, and it currently leads U.S.-based open-weights models on the Artificial Analysis Intelligence Index. However, while Inkling outperforms other open models in the U.S., top Chinese open-weights models still beat it on some key benchmarks. Thinking Machines prices access at $1.87 per million input tokens and positions Inkling mainly as a base model for fine-tuning, rather than claiming it as the most powerful AI available.
Why it matters
Inkling represents a major step in U.S. open AI models by pushing closer to the scale of China’s largest open models. The 975 billion parameter size is big enough to handle complex multimodal tasks, making it more useful for companies and developers looking for flexible AI they can customize. By offering open weights and a clear positioning as a fine-tuning base, Thinking Machines strengthens the ecosystem of accessible, adaptable AI outside closed proprietary models. Its pricing model also makes it competitive for startups and research teams that need scalable AI without the expense of top-tier closed models.
Still, the model’s performance gap with Chinese competitors shows that U.S. research and infrastructure face pressure to close the gap on scale and task-specific capabilities. Inkling puts more spotlight on the importance of fine-tuning strategies and model accessibility given how raw size no longer guarantees dominance. The fact that this effort comes from a top OpenAI alumnus signals that talent and ideas are actively flowing to new hubs attempting to decentralize AI development.
What to watch next
Watch for how well Thinking Machines’ fine-tuning ecosystem around Inkling develops. The practical power of these models depends heavily on accessible tooling and quality training data. The pricing at under two dollars per million tokens could attract startups and labs looking for alternatives to cloud-based closed models, but market traction will depend on ease of use and integration. Also track responses from Chinese open AI labs, which currently hold some performance edges. Any improvements in Inkling or new iterations might pressure other open labs in the U.S. to innovate faster or scale up. Finally, follow whether Inkling can carve a niche among builders who prioritize open weights paired with competitive quality over sheer model scale.
AI Quick Briefs Editorial Desk
