Thinking Machines' 26.4B Inkling 2 Mini Matches 975B Giants at 17x Compute Efficiency

Third-party benchmarking by Vals AI shows Thinking Machines' Inkling 2 Mini, a 26.4B-parameter model small enough to run on a MacBook, matches Claude Opus 5.5 on cybersecurity benchmarks at under a quarter of the cost, achieving roughly 17x compute efficiency after 128 days of training.

2026-11-21 ~ 2026-11-21 · 2 related posts