Inkling AI Scores 97.1% on AIME 2026 Benchmark

Mira Murati’s Thinking Machines Lab has launched Inkling, its first general-purpose AI model, which scored 97.1% on the AIME 2026 benchmark. This open-weight model works with text, image, and audio inputs. It has a 1 million token context window. Inkling is a Mixture-of-Experts model with 975 billion total parameters. It uses 41 billion parameters for each task, which helps reduce computing needs. The model was trained on 45 trillion tokens from text, images, audio, and video. While it accepts various inputs, it only generates text responses. Thinking Machines says Inkling is designed to be broad, trained across many tasks like reasoning, coding, and vision. It competes closely with other major open-weight models, though some closed models still lead in certain areas. This new model offers developers a broad base for fine-tuning their own AI applications.