← All model drops
Live drop
Inkling
Thinking Machines LabDropped Jul 15, 2026
Native multimodal MoE with controllable thinking effort for efficient long-horizon agentic coding and tool-use workflows.
About this release
Live drop
Thinking Machines LabDropped Jul 15, 2026
Native multimodal MoE with controllable thinking effort for efficient long-horizon agentic coding and tool-use workflows.
About this release
Inkling is Thinking Machines Lab's first open-weight multimodal model: a 975B-parameter (41B active) decoder-only MoE transformer pretrained on 45T multimodal tokens. It natively accepts text, image, and audio inputs, supports up to 1M-token context, and features controllable test-time compute for balancing performance against token cost. Released under Apache 2.0 with full weights on Hugging Face, it serves as a customizable generalist foundation for agentic applications, coding harnesses, and fine-tuning on the Tinker platform. A lighter Inkling-Small preview (12B active) was released alongside it.
Read the official Thinking Machines Lab launch pageFirst builds incoming — this lane fills in automatically as people start building with Inkling. Know one? Submit the URL.