PANews 1月21日消息，据量子位报道，DeepSeek在GitHub更新的FlashMLA代码中首次出现“MODEL1”名称，覆盖114个文件中28处提及，且与现有版本V32（DeepSeek-V3.2）并列，暗示MODEL1为下一代新架构模型。代码差异显示该模型在KV缓存布局、稀疏性处理及FP8解码等方面进行了优化，或将在春节前后正式发布。结合近期公开的mHC残差连接机制与Engram记忆模块，MODEL1有望整合多项自研创新。
DeepSeek新模型MODEL1代码曝光，疑为全新架构
Favorite
Share
Disclaimer: This article is copyrighted by the original author and does not represent MyToken’s views and positions. If you have any questions regarding content or copyright, please contact us.(www.mytokencap.com)contact
About MyToken:https://www.mytokencap.com/aboutusArticle Link:https://www.mytokencap.com/news/556093.html
More exciting content is available on
X(https://x.com/MyTokencap)or join the community to learn more:MyToken-English Telegram Group
（https://t.me/mytokenGroup）
X(https://x.com/MyTokencap)or join the community to learn more:MyToken-English Telegram Group
（https://t.me/mytokenGroup）
Previous:顾景辞：1.21比特币/以太坊操作策略附行情分析
Related Reading
Bitcoin’s Quantum Threat Is Evolving From a Sci-Fi Concept Into a Tangible, Systemic Economic Headwind
龙虾看币圈2026-10-31 05:05:14
Binance Hit by $6.3 Billion Hack, Pledges Full Compensation
龙虾看币圈2026-10-31 05:01:41
Bitcoin ETFs Snap a Four-Day Inflow Streak With $409M in Post-Hack Outflows
U.S. spot Bitcoin ETFs posted their first outflows after the Binance hack, but the selling was conce...
blockchainreporter2026-10-31 04:42:19