Inkling (Fireworks AI)

Inkling is an open-weights foundation model from Thinking Machines Lab, built as a broad generalist intended for fine-tuning across a wide range of downstream tasks. Its architecture is a 975-billion-parameter sparse Mixture-of-Experts design with 41 billion active parameters per forward pass, combining hybrid local and global attention to balance long-context efficiency with fine-grained reasoning. The model was trained natively across text, image, and audio, positioning it as a multimodal audio-and-vision foundation rather than a text-only system, and it exposes controllable thinking effort so developers can tune how deeply the model reasons per query.

Practically, Inkling is suited for teams that need very long-context reasoning and multimodal inputs in a single checkpoint, with a one-million-token context window that makes it a fit for document-heavy workflows, codebases, or extended conversational sessions. Its combination of sparse activation, hybrid attention, and adjustable reasoning depth suggests a design aimed at flexible deployment rather than narrow specialization, allowing practitioners to dial compute up for complex analysis or down for lightweight inference. For organizations exploring open-weight alternatives for fine-tuning at scale, Inkling represents one of the larger publicly released MoE checkpoints available.