AI 导读
Cognition 发布新的快速高效编码模型 SWE-2 Flash，官方称其在主要评测上超过所有美国开源权重模型，包括参数量 3 倍于它的模型。作者指出它在部分基准上接近更大的 SWE-2，价格为其五分之一，token 生成速度约 3 倍；该模型基于 Thinking Machines Lab 的 Inkling-Small（276B 总参数、12B 激活参数）训练，官方还计划通过安全项目与合作伙伴和研究者共享权重。
正文
Cognition’s SWE-2 Flash looks like a meaningful efficiency release for coding agents: it approaches the larger SWE-2 on some benchmarks at one-fifth the price and roughly 3× the token generation speed.
Even more interesting: its based on Mira Muartis thinking machine labs inkling-small.
Introducing SWE-2 Flash, our new fast & efficient coding model. On leading evals, it outperforms all American open-weight models, even ones 3x larger. To balance transparency and cyber risks, we are building a program to securely share the weights with partners & researchers.在 X 查看被引用的帖子
来源：Chubby♨️ · x.com