Thinking Machines 发布开源小模型 Inkling 2 Mini,笔记本可跑
Mira Murati 团队这次的 Inkling 2 Mini 有意思:中间层每个 token 跑三遍,用更少 token 就能完成推理。20GB 显存笔记本能跑,DeepSWE 48.2% 拿下百亿级以下开源模型第一。想在本地折腾编程推理的可以试试。
Thinking Machines 发布了开源权重模型 Inkling 2 Mini,参数量 26.4B(激活参数 3.4B),显存占用约 20GB,可以直接在笔记本上运行。
它的核心技术是循环 Transformer 架构:中间层每个 token 重复执行三次,让模型在内部完成更多推理,从而用更少的输出 token 达到同等效果。按 Thinking Machines 的数据,Inkling 2 Mini 在 DeepSWE v1.1 上拿到 48.2%,是 100B 参数以下开源模型中的最好成绩。
这款模型由 Mira Murati 团队推出,延续了其做小体积开源模型的路线。对想在本地跑编程推理任务的开发者来说,3.4B 的激活参数意味着消费级硬件即可承载,且推理成本比同效果的大模型更低。
This is such a cool release: Thinking Machines released Inkling 2 Mini, an open-weights model small enough to run on your laptop.
It's a looped transformer. The middle layers run three times per token, so it reasons more internally and needs fewer tokens to get there.
- 26.4B params (3.4B active), ~20GB memory - 48.2% on DeepSWE v1.1, the best of any open model under 100B, according to Thinking Machines
Ngl, @miramurati and team are building super impressive small open weight models. I love it!
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Inkling 2 Mini, Thinking Machines, open-source models, reasoning models.
Source attribution: curated from kimmonismus (@kimmonismus). Cross-source corroboration: 6 independent sources have covered this development (including Artificial Analysis, Merve Noyan, Thinking Machines Lab), confirming industry-wide signal strength.