N環球 N GLOBE
US Technology

Thinking Machines releases open model Inkling 2 Mini that runs locally on a laptop

· N GLOBE

Thinking Machines Lab has released Inkling 2 Mini, an open-weights model with 26.4 billion parameters in a roughly 20GB package that can run locally on a 32GB Mac or a single DGX Spark, under an Apache 2.0 licence.

Thinking Machines Lab, led by Mira Murati, released the open-weights model Inkling 2 Mini on November 20, pitching it as a model that runs locally on a laptop: 26.4 billion total parameters in a roughly 20GB package, runnable on a 32GB Mac or a single DGX Spark, under an Apache 2.0 licence.

A looped architecture: think more with fewer parameters

Inkling 2 Mini uses a looped transformer architecture: of its 24 layers, the middle 12 are run three times for every token, letting the model "think several times over" with fewer parameters. It activates 3.4 billion parameters per token, or 5.4 billion effective parameters counting the loops, with a context length of 1 million tokens.

Performance and licensing

Inkling 2 Mini scored 79.2% on SWE-bench Verified, leading agentic coding among open models under 100 billion parameters. Earlier this month the company open-sourced the weights of Inkling 2 Small; Mini is the lighter new member of the family, aimed at developers who need to deploy locally or in constrained environments.

💬 Comments

No comments yet. Be the first!