Inkling 2 Mini matches Qwen 3.8 27B on SVG quality, nearly 3x faster locally
LiTianleli · x · 2026-11-22
Thinking Machines' Inkling 2 Mini is a 26.4B MoE model with only 3.4B active parameters per token, using 20GB total so it runs on a 32GB Mac. It scored 74.58% on CyberBench, matching Claude Opus 5.5 (CVP) at less than a quarter of the cost per run.
A blogger tested it against Qwen 3.8 27B on a 32GB M5 Pro MacBook using the classic "pelican riding a bicycle" SVG prompt. Visual quality was close, with Inkling slightly sharper; the real gap was speed and efficiency:
- Qwen 3.8 27B (Q4): 18–20 tok/s, 2,800 tokens, a bit over 2 minutes
- Inkling 2 Mini: 55–65 tok/s, 1,300 tokens, about 22 seconds
Verdict: for local Mac use, Inkling 2 Mini is the easy choice.
Related event: Inkling 2 Mini Beats Qwen 27B Locally on a Single RTX 4060(2 posts)→
More from Models
- Google AI Mode now enforces usage limits, users surprised by rate-limit popup — elvano124 · 2026-11-22
- Sol 6.2 now refuses reverse engineering that Sol 6.1 happily did for a month — cardak98 · 2026-11-22
- Anthropic's record ~$81B IPO, GPT-6.2 pricing, Qwen4-Max and Copilot Autopilot: week in review — ksraj1001 · 2026-11-22
- Reddit rant: Frontier AI labs are training models to imitate lazy engineers — RossPeili · 2026-11-22
- MBZUAI open-sources PAN 2, a 29B-parameter general-purpose world model — LinusEkenstam · 2026-11-22
- Inkling Mini turns out surprisingly fluent in Russian, sparking data recipe questions — teortaxesTex · 2026-11-22