Inkling 2 Mini tops efficiency charts: 26B looped transformer at $25.80 per correct answer
TheZachMueller · x · 2026-11-22
p0's new benchmarks crown Inkling 2 Mini as the most efficient model tested: #1 on cost-vs-intelligence, #9 overall (78.4), #1 among open-weights, lowest cost per correct answer ($25.80), and the biggest accuracy gain from web search. The looped transformer runs its middle layers three times per token, letting a 26B model reason more without generating more tokens. Mira Murati says it runs on a laptop and its thought process remains monitorable at this scale; a variable-loop research model is also being released.
Related event: Inkling 2 Mini Tops Cost-Performance Rankings as 26B Open-Weight Model(2 posts)→
More from Infra
- A Hands-On GPU Profiling Starter Path: Time PyTorch Attention with Stanford CS336 Assignment 2 — stanfordnlp · 2026-11-22
- 42 Sellers, 5x Price Spread: Open Models Enrich Big Inference Providers, Not Startups — AccBalanced · 2026-11-22
- 425MB Qwen3.5-0.8B decision model on CPU: custom 4-bit quant loses just 0.18pp, 2x faster — Danmoreng · 2026-11-22
- Gemini 4 Neon autonomously writes GPU kernels, 5.9x faster prefill on AMD hardware — sebkrier · 2026-11-22
- SenseNova-Loop-9B Hits 1,300+ tok/s on a Laptop RTX 5090, Benchmarked vs Qwen 27B — tebjan · 2026-11-22
- Intel's 350W D270 claims 5.9x lead over RTX PRO 6000 — but the fine print reveals benchmark games — StefanoGogioso · 2026-11-22