Thinking Machines Lab has released Inkling 2 Mini, a new looped transformer model designed for local AI setups. This Mixture-of-Experts model features approximately 26 billion total parameters and 3 billion active parameters per token, utilizing a novel looping mechanism in its middle layers to enhance capacity and reasoning. Inkling 2 Mini supports a 1 million token context size, accepts both image and text inputs, and is available in mixed-precision formats for efficient local deployment. AI
IMPACT This model's efficient architecture and local deployment capabilities could accelerate the adoption of powerful AI tools on consumer hardware.
RANK_REASON New model release from a research lab with novel architecture (looped transformer). [lever_c_demoted from frontier_release: ic=2 ai=1.0]
- Hugging Face
- Inkling 2 Mini
- Inkling 2 Small
- llama.cpp
- MacBook Pro
- Mlx
- MXFP8
- Nanbeige
- NVFP4
- Qwen 3.8-27B
- SGLang
- Thinking Machines Lab
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →