Thinking Machines releases Inkling 2 Mini, an open-weights model for local coding
Thinking Machines releases Inkling 2 Mini, an open-weights model for local coding
Thinking Machines has released Inkling 2 Mini, an open-weights transformer with a cyclic architecture that can run locally on a Mac and was used to build a flight simulator offline.
Thinking Machines released Inkling 2 Mini, an open-weights model available on Hugging Face.
The model is built on a cyclic transformer architecture: middle layers process each token three times. This allows it to perform more internal computation while using fewer tokens to solve tasks.
Key specs:
- 26.4B parameters, with 3.4B active.
- Requires about 20 GB of memory to run.
- Scores 48.2% on
DeepSWE v1.1— according to Thinking Machines, the best result among open models with fewer than 100B parameters.
The Thinking Machines team took Inkling 2 Mini on a plane and asked the model to create a flight simulator.
Everything ran locally on a Mac without an internet connection.
The developers demonstrated that the model can be used for programming even without network access.
Why it matters
AnalysisInkling 2 Mini shows that open-weights models are becoming practical for offline coding workflows on consumer hardware, reducing dependence on cloud APIs.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
Meta Llama (The Premier Open Standard for Artificial Intelligence)
The flagship series of open language models from Meta (Llama 3, 3.3). Available for free download, modification, and local deployment on personal servers without restrictions or censorship.
Llama Family (Meta Llama)
A series of foundational open language models from Meta (Llama 3, 3.1, 3.3) that have become the industrial standard for the Open Weights ecosystem, local AI, and enterprise fine-tuning.
Thinking Machines Releases Looped Transformer Inkling 2 Mini
Thinking Machines, the startup founded by Mira Murati, has released Inkling 2 Mini, a local AI model built on a looped transformer architecture that repeats middle layers to deepen reasoning without increasing parameter count.
Google unveils Gemini 4 Neon: model outperforms Claude Opus 5.5 and GPT-6.2 Sol on coding and agentic tasks
Google has introduced Gemini 4 Neon, a new AI model that reportedly beats Claude Opus 5.5 and GPT-6.2 Sol on most tests, especially in programming, reasoning, and agent workflows.