Thinking Machines ships Inkling 2 Mini: 26B MoE that runs on 16GB RAM, tops its size class
SonglinYang4 · x · 2026-11-21
Thinking Machines released Inkling 2 Mini, an Apache 2.0 MoE model with 26.4B total params (5.4B active) using a Looped Transformer architecture with 1M context, aimed at coding, agentic, and vision workloads.
- Benchmarks: 48.2% on DeepSWE v1.1, 77.9% on SWEBench Pro v2 (Hard), 79.2% on SWEBench Verified, 79.0% on BrowseComp — best in its size class; 20.1 on AA Index
- Local deployment via Unsloth: GGUF quants from 9GB (1-bit) to 24GB (8-bit lossless); runs on a 16GB-RAM laptop
- Released in response to feedback that Inkling 2 Small (594B) was too big; available on Unsloth and llama.cpp
More from coding & agent
- Full workflow fixes Fable 5.2 video audio entirely inside Claude — AIwithGhotai · 2026-11-21
- Claude Opus 5.5 drives Onshape CAD via computer use, with surprisingly delightful results — adhamelarabawy · 2026-11-21
- Inkling 2 Mini runs fully offline on a MacBook, builds an airplane simulator in minutes — LiTianleli · 2026-11-21
- Inkling 2 Mini runs locally in 16GB RAM: 5.4B-active MoE tops its size class — simonguozirui · 2026-11-21
- Replit engineer: 'I no longer need anything locally for my job' — amasad · 2026-11-21
- Replit Agent lands in Microsoft Teams to turn chats into action — amasad · 2026-11-21