signal量子位（公众号）zh source2026-09-21
一张3090就能跑！全栈国产模型，把AI办公搬到企业本地
China Telecom open-sourced Xing4.0-29B-A4B, a fully domestic (Chinese) MoE code-agent model with 29B total parameters but only ~4B activated per inference, enabling local deployment on a single RTX 3090 after 4-bit quantization (~15GB VRAM, a 75% cut from FP16). Trained on Huawei Ascend 910C with MindSpore/MindFormers, it supports 256K - 512K context and handles enterprise tasks like reviewing 200-page bid documents in ~20 minutes or converting contract spreadsheets into dashboards, with a 90%+ success rate in private-deployed customer-service workflows; it's available on Hugging Face, GitHub, and Gitee, and reached #4 on HF trending.
- why now
- 中国电信今日开源Xing4.0，3090可跑，冲上HF榜第四
- topic
- AI Tech & New Models
- source
- 量子位（公众号）
#MoE
score
score 7 out of 10. 0-10: how dense the facts are, multiplied by how much you can do with them after reading. 8+ means the topic's evidence bar is met: benchmarks and availability for a new model, amount and investors for a funding round, revenue figures for a solo-money story. Below 5 an item does not enter the digest. A press release scores 3 or less, a reprint loses 2, anything older than 14 days loses 1, a headline that misleads loses 3.
read the source