Thinking Machines releases Inkling: open-weight multimodal MoE with strong benchmarks, controllable thinking, and enterprise customization
Key Questions
What is the Inkling model released by Thinking Machines?
Inkling is an open-weight multimodal MoE model with 975B total parameters and 41B active parameters, supporting text, image, and audio. It was built from scratch with a controllable thinking effort mechanism.
Who founded Thinking Machines and what is their focus with Inkling?
The company was founded by Mira Murati, John Schulman, and others. Inkling emphasizes enterprise customization, low cost, and resistance to censorship as a Western alternative to Chinese open-weight models.
What license and availability options exist for Inkling?
It comes with an enterprise-friendly license and is available via GGUF quantized versions on Unsloth. The model can also be accessed through Tinker.
How does Inkling compare to other open-weight models?
It delivers competitive benchmarks against top closed models while challenging Chinese dominance in the open-weight space. The model prioritizes customization and cost-performance.
What makes Inkling suitable for enterprise use?
Its license supports customization, and the model resists censorship. Quantized versions enable efficient deployment for agentic workloads.
Thinking Machines (founded by Mira Murati, John Schulman, and others) released Inkling, an open-weight multimodal MoE model (975B total, 41B active) covering text, image, and audio. It achieves competitive benchmarks against top closed models, with a novel controllable thinking effort mechanism and enterprise-friendly license. The model is built from scratch (not fine-tuned) and emphasizes customization and resistance to censorship. GGUF quantized versions are available via Unsloth. This challenges Chinese open-weight dominance and provides a Western open-source alternative with strong cost-performance.