Mira Murati's Thinking Machines Lab Releases First Open-Weight Model Inkling

Inkling is a Mixture-of-Experts (MoE) Transformer model with 975B total parameters (41B active), trained on 45 trillion tokens of text, image, audio, and video data, released under Apache-2.0. The model card is brief, and training data documentation only states use of public and third-party datasets. Thinking Machines Lab also promised a smaller Inkling-Small variant (276B total, 12B active), still in testing. Inkling is not a frontier model but a strong base model designed for fine-tuning via the Tinker platform, offering multimodal capability, efficient inference, and direct fine-tuning on Tinker. It competes with open-weight models like NVIDIA Nemotron and Gemma 4, injecting new vitality into the US open-source ecosystem. Users can call Inkling via the Thinking Machines API; example curl commands and SVG generation show its multimodal reasoning (though generated SVG misidentified a pelican as a 'stork or seagull').