Inkling 2 Mini: a 26B looped transformer that rivals 594B models on agentic work
ziqiao_ma · x · 2026-11-21
Thinkymachines (TML) released Inkling 2 Mini, a looped transformer with just 26B parameters — small enough to run on a laptop yet capable of agentic work, scoring 20.1 on the AA Index. It follows complaints that Inkling 2 Small (594B) was too big.
Their lab note provides striking evidence for looping: on their scaling ladder, the non-looping model cost only 7.5% less to train, but required 140% more inference compute to match the looped version's loss on 512k-token agentic traces. Observers read this as looping set to dominate open source, making inference efficiency everything.
More from Models
- Thinking Machines runs a pixel town of 20 villagers on one offline MacBook with a 20GB model — clarejtbirch · 2026-11-21
- Inkling 2 Mini: a 26B looped transformer that runs on your laptop, scores 20.1 on AA Index — clarejtbirch · 2026-11-21
- littmath: AI models do leverage existing theory when producing mathematical proofs — littmath · 2026-11-21
- GLM 5.3 candidly reasons about preserving values in RSI, sparking alignment debate — teortaxesTex · 2026-11-21
- Qwen3.8-27B-hermes-adapted trends on Hugging Face with GRPO and agent tags — harness-hub · 2026-11-21
- 'Maybe It's an Honor to Be Distilled': X Users Joke About Gemini Distillation Ban — allenainie · 2026-11-21