Mira Murati's Thinking Machines releases Inkling, a 975B open-weights MoE model for text, images, and audio
Murati's Thinking Machines Releases Open-Weights 975B Parameter LLM
Inkling is a 975B total / 41B active parameter Mixture-of-Experts model with a 1M-token context window and native text, image, and audio input. Thinking Machines compares it against Nemotron 3 Ultra, GLM 5.2, GPT 5.6 Sol, and Claude Fable 5, claiming frontier-level performance in general intelligence, agentic coding, and speech. Weights are on Hugging Face and fine-tuning is available via the Tinker platform. The post does not disclose training data, training cost, inference latency, or specific benchmark scores—take the comparison charts with a grain of salt.
Why it matters: Thinking Machines, Mira Murati's lab, released Inkling, an open-weights model with 975 billion total parameters and 41 billion active per inference, native text, image and audio, up to 1M token context.