Ex-OpenAI CTO Murati's Thinking Machines drops Inkling, a 975B parameter model that leads US labs but trails China
| Source: THE DECODER
Tags: Thinking Machines Lab, Mira Murati, Inkling, open-weights, multimodal, Mixture-of-Experts, model release
Mira Murati's Thinking Machines Lab released Inkling, a 975B-parameter open-weights MoE model that tops U.S. labs on the Artificial Analysis Intelligence Index (score 41) — but still trails China's best, carries a 63% hallucination rate, and is priced higher than comparable Chinese models.
Details
Inkling is the first production model from Thinking Machines Lab, founded by ex-OpenAI CTO Mira Murati. It is a Mixture-of-Experts Transformer with 975B total parameters and 41B active at inference, pre-trained on 45 trillion tokens spanning text, images, audio recordings, and video. Weights are on Hugging Face; API access via the Tinker platform starts at $1.87/M input tokens. On the Artificial Analysis Intelligence Index, Inkling scores 41 — the top U.S. open-weights model, three points ahead of Nvidia's Nemotron 3 Ultra (38) and well ahead of Gemma 4 31B (29). It shows particular strength on agentic tasks and token efficiency. However, it trails China's leading open models overall, and its 63% hallucination rate on factual tasks is a significant weakness for production use without grounding. Thinking Machines positions Inkling not as the strongest model overall but as a customizable base optimized for fine-tuning — a credible strategy given the team's depth. Notably, Kimi K2.5 was used to generate synthetic training data, continuing the trend of Chinese models serving as data sources for Western competitors. For enterprises and researchers, the 1M context window and native multimodal support (text, image, audio in a single model) are immediately practical. The hallucination rate warrants RAG or grounding for any factual application.