Thinking Machines Lab releases Inkling, a multimodal MoE model

Inkling accepts text, image, and audio inputs and outputs text, with controllable inference effort balancing reasoning depth, token usage, and latency. It is available on Together AI's inference platform, served via an optimized FlashAttention-4-based kernel, and post-trained for scientific reasoning, coding, and agentic workflows.