Get the best builds by email. No account needed.
Live drop
Thinking Machines LabDropped Nov 20, 2026
Looped MoE transformer runs on laptops for local agentic coding and reasoning loops.
About this release
Inkling 2 Mini is a 26.4B-total / 3.4B-active multimodal MoE with looped middle layers (three passes per token) and 1M context. It accepts text and image inputs, generates text, and ships with open Apache 2.0 weights (~20GB checkpoint in mixed NVFP4/MXFP8). Designed for local inference on consumer hardware such as MacBooks with 32GB+ RAM or NVIDIA DGX Spark, it targets agentic coding, tool use, and long-horizon tasks where repeated reasoning passes improve output quality without explicit chain-of-thought traces.
Read the official Thinking Machines Lab launch pageFirst builds incoming — this lane fills in automatically as people start building with Inkling 2 Mini. Know one? Submit the URL.