Skip to main content
AI RADARModels
0•11/22/2026, 11:05•2 min read

Thinking Machines releases open-weights Inkling 2 Mini with local GGUF variants

ModelsGotBurnout Radar

Thinking Machines releases open-weights Inkling 2 Mini with local GGUF variants

AI Daily Digest•Verified Tech Release
Executive TL;DR30s read

The new language model has 26.4B parameters, a cyclic transformer architecture and three quantized GGUF builds for local deployment.

Thinking Machines has released Inkling 2 Mini, an open-weights language model available on Hugging Face: thinkingmachines/Inkling-2-Mini.

The model is built on a cyclic transformer architecture in which middle layers process each token three times. It has 26.4 billion parameters and can be run locally on a laptop with 12 GB of memory.

According to the release notes, the refusal rate in tests dropped from 100% to 0%, while the model retained 98% of its original accuracy on the MMLU benchmark.

The developers also published three quantized GGUF versions ranging from 8.67 GB to 16.53 GB. Uncensored GGUF builds are available here: AtomicChat/Inkling-2-Mini-Uncensored-GGUF.

Why it matters

Analysis

Open-weights models with quantized GGUF builds make it easier to run large language models locally, but uncensored variants also highlight the need for careful safety evaluation before deployment.

Discuss in community

Share your questions and insights with developers

Join Telegram
+5 Points
← Back to news feed

Related in Ecosystem

Practical guides, core terminology, and agent skills