Models New
Thinking Machines releases Inkling 2 Mini, a 26-billion-parameter open-weights model small enough to run on a 32GB Mac
Published:
Thinking Machines Lab released Inkling 2 Mini on Friday, an open-weights model under the Apache 2.0 licence that it says is small enough to run on a laptop. The weights are on Hugging Face, the model is on the company's Tinker platform, and Ollama lists it for Macs with 32GB or more of memory. It was first posted to Hacker News at 21:12 Paris time on Friday, after the previous afternoon edition had closed, and the morning edition did not carry it.
Inkling 2 Mini is a mixture-of-experts model with 26.4 billion total parameters and 3.4 billion active per token. It is a looped transformer: the middle 12 of its 24 layers run three times for every token, which the company says lets it reason more without writing more tokens. Most of its weights were trained and ship in 4-bit NVFP4, for a memory footprint of about 20GB.
Thinking Machines says the model leads every open-weights model under 100 billion parameters on DeepSWE v1.1 (48.2%) and Vals Index v2.2 (46.5%). It also says the model is weaker on multi-turn chat and on instructions with many constraints, where it trails Qwen3.8-27B. Artificial Analysis, which tests models independently, placed it at 20 on its Intelligence Index, level with Qwen3.8 27B, at $0.59 per task against $3.53.
The lab also released a research variant with a variable number of loops, used to test whether looping hides reasoning from people reading a model's chain of thought. It reports that the chain of thought got shorter with more loops but stayed legible, and that the model did not score higher on tests of agentic sabotage as loops were added.
Caveat. The benchmark results are Thinking Machines' own, except the Artificial Analysis index. The release page gives only the date; the time comes from the first Hacker News post.
Sources