Thinking Machines releases Inkling 2 Mini, a 26B looped MoE that runs on a 32GB Mac
Thinking Machines Lab has released Inkling 2 Mini, an open-weight looped transformer with 26.4B total parameters and 3.4B active parameters (5.4B effective per token) that reuses its middle layers three times rather than adding weights. The checkpoint is roughly 20GB under Apache 2.0, supports up to 1M-token context, and is listed to run on Apple silicon with 32GB+ memory, a single NVIDIA DGX Spark, or GPUs with 32GB+ VRAM. Thinking Machines self-reports 79.2% on SWE-bench Verified (bash only) and a 20.1 Artificial Analysis Index score tied with Qwen3.8-27B, while acknowledging weaker knowledge recall and multi-turn constraint following. GGUF builds are available via Unsloth and llama.cpp, with serving support across vLLM, SGLang, Ollama and Transformers.
Why it matters: It gives developers a concrete, locally runnable open-weight coding model with published architecture details, hardware requirements and self-reported benchmark trade-offs.