Primary sourceDecoder
View sourceEvent ·Multi-source
Thinking Machines 开源 Inkling 2 Mini:循环推理省 token,笔记本可跑
Thinking Machines 发布开源模型 Inkling 2 Mini,采用循环 transformer 架构,24 层中让 12 层对每个 token 重复执行三次,从而在内部延长推理过程而不产生额外输出文本。在 Artificial Analysis 的 Intelligence Index 基准上,它与 Qwen3.8 27B 打平,成本约低六倍。由于总参数规模较小,该模型可以在普通笔记本上本地运行。循环机制的含义是把算力花在重复利用层参数上,而不是堆叠更深的网络或生成更长的思维链,为低算力场景提供了一条不同的推理路径。对开发者来说,这意味着在不依赖云端 GPU 的情况下获得接近更大开源模型的推理表现,同时显著降低 token 开销。
Current read
Thinking Machines 放出的 Inkling 2 Mini 用循环重复 12 层的办法做内部推理,不靠长思维链,成本只有 Qwen3.8 27B 的约六分之一还打平成绩。想在笔记本上本地跑推理模型的开发者可以重点关注这个架构。
11 sources47° AI heatLast updated 2026/11/21 10:33:02
Evidence
11 sourcesRelated source 1Artificial Analysis
View sourceRelated source 2Julien Chaumond
View sourceRelated source 3kimmonismus
View sourceRelated source 4lmarena.ai
View sourceRelated source 5marktechpost
View sourceRelated source 6Clement Delangue
View sourceRelated source 7Merve Noyan
View sourceRelated source 8Thinking Machines Lab
View sourceRelated source 9vLLM
View sourceRelated source 10ollama
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.