Inkling-2-Mini: Thinking Machines' 26.4B looped-transformer MoE runs in 16GB RAM
simonguozirui · x · 2026-11-21
Thinking Machines released Inkling-2-Mini, an Apache 2.0 MoE model using a looped transformer architecture that cycles some layers multiple times — 26.4B total params with 5.4B active (including loops). It scores 48.2% on DeepSWE v1.1, the best for its size class.
UnslothAI produced UD3.1 quants that run on 16GB RAM via Unsloth desktop and llama.cpp. Their benchmark chart shows 7-domain mean top-1 accuracy climbing from 89.5% (IQ1S, 8GB GGUF) to 98% (Q4KXL, 16GB+), with small quants like IQ1M retaining 90%+ accuracy.
Related event: Inkling 2 Mini lands on Ollama, runs locally in 16GB of RAM(5 posts)→
More from Models
- KOL endorses Google AI Ultra: screenshot shows up to 5x Gemini limits vs Pro — kimmonismus · 2026-11-21
- Claude's claimed Riemann proof mocked: 250,000 pages of unreviewed Lean, author calls it 'slop' — GaryMarcus · 2026-11-21
- First side-by-side test: Gemini Neon beats Opus 5.5 and peers at building a dataset viewer — zacharynado · 2026-11-21
- Frontier models can decode KV-cache side channels in context; researcher shaken — davidad · 2026-11-21
- GPT-6.2 Astra tops RoboColosseum real-world robot Elo leaderboard in one day — DJiafei · 2026-11-21
- OpenAI staff coordinate vagueposts teasing a mystery model: '[Redacted] is very good at [redacted]' — Kyrannio · 2026-11-21