Murati's 975B Inkling leads US open-weights but hallucinates 63% of the time
mira murati's thinking machines labs launched inkling, a 975b parameter open-weights model leading us benchmarks on agentic tasks despite a 63% hallucination rate and higher costs than chinese rivals.
thinking machines labs released inkling, a 975-billion parameter mixture-of-experts transformer with 41 billion active parameters, pre-trained on 45 trillion tokens. it scores 41 on the artificial analysis intelligence index, topping us rivals like nemotron 3 ultra and beating kimichinese models on specific agent tasks like banking. however, the model exhibits poor factual accuracy with a 63% hallucination rate and remains more expensive than comparable chinese alternatives, potentially limiting its enterprise deployment.