Inkling: A New Open-Weight 975B MoE with a Few Surprises

This article analyzes the surprise release of Inkling, a 975B-parameter sparse Mixture-of-Experts (MoE) open-weight LLM from Thinking Machines Lab. It compares Inkling's benchmarks against GLM-5.2, noting strengths in IFBench and SimpleQA Verified but weaknesses in reasoning and coding-agent tasks. The article details architecture: 41B active parameters, 1M token context, and unique features like small convolution layers for local token mixing, an additional RMSNorm after embedding, and learned input-dependent relative-position bias. It discusses token throughput considerations and positions Inkling as a versatile base model for fine-tuning via the Tinker platform.