{"id":85865,"topic":"ai","source":"新浪财经_金融信息服务商","title":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配 - 新浪财经_金融信息服务商","url":"https://finance.sina.cn/stock/jdts/2026-09-19/detail-iniskptr2203813.d.html?oid=800&vt=4&cid=76993&node_id=76993","url_hash":"e51a3f3251a21b0783c53eebf74fad50e5e2a26c","author":"","summary":"<a href=\"https://news.google.com/rss/articles/CBMisgFBVV95cUxQMl9jbjJENU1na1dNdXpwLWVjLXd0RGhTa2tVampBUXdDNFcwNmtJMXlfeElqaVFmMnFxVFE2SEdYbmZQa09xajVabVl5Q0sweW9VS0VoZUJtWXpLTUJnMm92cDdmbjRPM1FEZnplWVpUSnNHeUplMTcxSC1EU2stSmJHTFoxQWV2VzBjNzdyZXRJN21qMm45dElUbFo1Sm5wTVZsb1Z1bld5eF9QY2JMZldn?oc=5\" target=\"_blank\">国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配</a>&nbsp;&nbsp;<font color=\"#6f6f6f\">新浪财经_金融信息服务商</font>","content":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配\n（来源：快科技）\n快科技9月19日消息，中电信人工智能科技有限公司日前正式发布新一代星辰大模型Xing4.0-29B-A4B。沐曦股份宣布，依托自研MXMACA软件栈，曦云C系列GPU率先完成对该模型的Day 0适配，实现\"上线即适配\"。\nXing4.0-29B-A4B是面向智能体时代打造的轻量级代码智能体大模型，采用MOE架构，总参数量29B，激活参数仅4B，原生支持256K上下文，可扩展至512K。\n该模型是国内首个基于国产算力与国产框架完成训练、面向复杂工程任务深度优化的百亿参数大模型，聚焦任务规划、工具调用和自主执行等能力。\n性能方面，Xing4.0-29B-A4B在SWE-bench Verified上得分75.0，Terminal-Bench 2.1得分57.5，均超过同尺寸竞品。\n在SuperCLUE智能体能力评测中，该模型以93.52分位列第三，与两款头部Qwen模型差距不足1分。4-bit量化后显存占用降至15GB，RTX 3090、RTX 4090等消费级显卡即可本地运行。\n沐曦此次Day 0适配的核心支撑来自MXMACA软件栈。该软件栈全链路覆盖底层驱动、MXCC编译器、算子深度适配及主流训练/推理框架对接，可将传统模型适配周期从数周压缩至小时级。\n目前MXMACA已兼容40余种AI框架，适配超1000款模型，全量支持PyTorch 2.8的2410个GPU算子。\n自2025年12月以来，沐曦已完成38个主流顶尖模型的Day 0适配，范围涵盖智谱AI、阿里千问、MiniMax、DeepSeek、阶跃星辰、腾讯混元等头部厂商，适配速度、数量与生态广度均位居行业前列。","image_url":"https://n.sinaimg.cn/spider20260919/157/w600h357/20260919/0230-c11335d50ba17bc837254cef5dd038f5.jpg","lang":"zh","published_at":"2026-09-19T13:26:03+00:00","fetched_at":"2026-09-19T14:15:08+00:00","status":"read","starred":0,"extract_state":"ok","summary_auto":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配\n（来源：快科技）\n快科技9月19日消息，中电信人工智能科技有限公司日前正式发布新一代星辰大模型Xing4.0-29B-A4B。沐曦股份宣布，依托自研MXMACA软件栈，曦云C系列GPU率先完成对该模型的Day 0适配，实现\"上线即适配\"。\nXing4.0-29B-A4B是面向智能体时代打造的轻量级代码智能体大模型，采用MOE架构，总参数量29B，激活参数仅4B，原生支持256K上下文，可扩展至512K。\n该模型是国内首个基于国产算力与国产框架完成训练、面向复杂工程任务深度优化的百亿参数大模型，聚焦任务规划、工具调用和自主执行等能力。\n性能方面，Xing4.0-29B-A4B在SWE-bench Verified上得分75.0，Terminal-Bench 2.1得分57.5，均超过同尺寸竞品。\n在SuperCLUE智能体能力评测中，该模型以93.52分位列第三，与两款头部Qwen模型差距不足1分。4-bit量化后显存占用降至15GB，RTX 3090、RTX 4090等消费级显卡即可本地运行。\n沐曦此次Day 0适配的核心支撑来自MXMACA软件栈。该软件栈全链路覆盖底层驱动、MXCC编译器、算子深度适配及主流训练/推理框架对接，可将传统模型适配周期从数周压缩至小时级。\n目前MXMACA已兼容40余种AI框架，适配超1000款模型，全量支持PyTorch 2.8的2410个GPU算子。\n自2025年12月以来，沐曦已完成38个主流顶尖模型的Day 0适配，范围涵盖智谱AI、阿里千问、MiniMax、DeepSeek、阶跃星辰、腾讯混元等头部厂商，适配速度、数量与生态广度均位居行业前列。","cluster_id":3513420,"extract_retries":0,"extract_error":null,"contract_version":"news_item.v1","format_contract_version":"news_item_formats.v1","dedup_url":"https://finance.sina.cn/stock/jdts/2026-09-19/detail-iniskptr2203813.d.html?oid=800&vt=4&cid=76993&node_id=76993","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 749 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":749,"summary_length":749,"usable_text_length":749,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":749,"summary_length":749}},"news_item":{"id":85865,"canonical_url":"https://finance.sina.cn/stock/jdts/2026-09-19/detail-iniskptr2203813.d.html?oid=800&vt=4&cid=76993&node_id=76993","source_url":"https://finance.sina.cn/stock/jdts/2026-09-19/detail-iniskptr2203813.d.html?oid=800&vt=4&cid=76993&node_id=76993","title":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配 - 新浪财经_金融信息服务商","source_name":"新浪财经_金融信息服务商","author":null,"published_at":"2026-09-19T13:26:03+00:00","locale":"zh","topic":"ai","tags":[],"rss_summary":"<a href=\"https://news.google.com/rss/articles/CBMisgFBVV95cUxQMl9jbjJENU1na1dNdXpwLWVjLXd0RGhTa2tVampBUXdDNFcwNmtJMXlfeElqaVFmMnFxVFE2SEdYbmZQa09xajVabVl5Q0sweW9VS0VoZUJtWXpLTUJnMm92cDdmbjRPM1FEZnplWVpUSnNHeUplMTcxSC1EU2stSmJHTFoxQWV2VzBjNzdyZXRJN21qMm45dElUbFo1Sm5wTVZsb1Z1bld5eF9QY2JMZldn?oc=5\" target=\"_blank\">国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配</a>&nbsp;&nbsp;<font color=\"#6f6f6f\">新浪财经_金融信息服务商</font>","full_text":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配\n（来源：快科技）\n快科技9月19日消息，中电信人工智能科技有限公司日前正式发布新一代星辰大模型Xing4.0-29B-A4B。沐曦股份宣布，依托自研MXMACA软件栈，曦云C系列GPU率先完成对该模型的Day 0适配，实现\"上线即适配\"。\nXing4.0-29B-A4B是面向智能体时代打造的轻量级代码智能体大模型，采用MOE架构，总参数量29B，激活参数仅4B，原生支持256K上下文，可扩展至512K。\n该模型是国内首个基于国产算力与国产框架完成训练、面向复杂工程任务深度优化的百亿参数大模型，聚焦任务规划、工具调用和自主执行等能力。\n性能方面，Xing4.0-29B-A4B在SWE-bench Verified上得分75.0，Terminal-Bench 2.1得分57.5，均超过同尺寸竞品。\n在SuperCLUE智能体能力评测中，该模型以93.52分位列第三，与两款头部Qwen模型差距不足1分。4-bit量化后显存占用降至15GB，RTX 3090、RTX 4090等消费级显卡即可本地运行。\n沐曦此次Day 0适配的核心支撑来自MXMACA软件栈。该软件栈全链路覆盖底层驱动、MXCC编译器、算子深度适配及主流训练/推理框架对接，可将传统模型适配周期从数周压缩至小时级。\n目前MXMACA已兼容40余种AI框架，适配超1000款模型，全量支持PyTorch 2.8的2410个GPU算子。\n自2025年12月以来，沐曦已完成38个主流顶尖模型的Day 0适配，范围涵盖智谱AI、阿里千问、MiniMax、DeepSeek、阶跃星辰、腾讯混元等头部厂商，适配速度、数量与生态广度均位居行业前列。","excerpt":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配\n（来源：快科技）\n快科技9月19日消息，中电信人工智能科技有限公司日前正式发布新一代星辰大模型Xing4.0-29B-A4B。沐曦股份宣布，依托自研MXMACA软件栈，曦云C系列GPU率先完成对该模型的Day 0适配，实现\"上线即适配\"。\nXing4.0-29B-A4B是面向智能体时代打造的轻量级代码智能体大模型，采用MOE架构，总参数量29B，激活参数仅4B，原生支持256K上下文，可扩展至512K。\n该模型是国内首个基于国产算力与国产框架完成训练、面向复杂工程任务深度优化的百亿参数大模型，聚焦任务规划、工具调用和自主执行等能力。\n性能方面，Xing4.0-29B-A4B在SWE-bench Verified上得分75.0，Terminal-Bench 2.1得分57.5，均超过同尺寸竞品。\n在SuperCLUE智能体能力评测中，该模型以93.52分位列第三，与两款头部Qwen模型差距不足1分。4-bit量化后显存占用降至15GB，RTX 3090、RTX 4090等消费级显卡即可本地运行。\n沐曦此次Day 0适配的核心支撑来自MXMACA软件栈。该软件栈全链路覆盖底层驱动、MXCC编译器、算子深度适配及主流训练/推理框架对接，可将传统模型适配周期从数周压缩至小时级。\n目前MXMACA已兼容40余种AI框架，适配超1000款模型，全量支持PyTorch 2.8的2410个GPU算子。\n自2025年12月以来，沐曦已完成38个主流顶尖模型的Day 0适配，范围涵盖智谱AI、阿里千问、MiniMax、DeepSeek、阶跃星辰、腾讯混元等头部厂商，适配速度、数量与生态广度均位居行业前列。","extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 749 characters.","diagnostics_url":"/api/diagnose?url=https%3A//finance.sina.cn/stock/jdts/2026-09-19/detail-iniskptr2203813.d.html%3Foid%3D800%26vt%3D4%26cid%3D76993%26node_id%3D76993","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 749 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":749,"summary_length":749,"usable_text_length":749,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":749,"summary_length":749}}},"display_formats":["compact","card","full","digest_section","json"]},"daily_stack_record":{"title":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配 - 新浪财经_金融信息服务商","url":"https://finance.sina.cn/stock/jdts/2026-09-19/detail-iniskptr2203813.d.html?oid=800&vt=4&cid=76993&node_id=76993","summary":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配\n（来源：快科技）\n快科技9月19日消息，中电信人工智能科技有限公司日前正式发布新一代星辰大模型Xing4.0-29B-A4B。沐曦股份宣布，依托自研MXMACA软件栈，曦云C系列GPU率先完成对该模型的Day 0适配，实现\"上线即适配\"。\nXing4.0-29B-A4B是面向智能体时代打造的轻量级代码智能体大模型，采用MOE架构，总参数量29B，激活参数仅4B，原生支持256K上下文，可扩展至512K。\n该模型是国内首个基于国产算力与国产框架完成训练、面向复杂工程任务深度优化的百亿参数大模型，聚焦任务规划、工具调用和自主执行等能力。\n性能方面，Xing4.0-29B-A4B在SWE-bench Verified上得分75.0，Terminal-Bench 2.1得分57.5，均超过同尺寸竞品。\n在SuperCLUE智能体能力评测中，该模型以93.52分位列第三，与两款头部Qwen模型差距不足1分。4-bit量化后显存占用降至15GB，RTX 3090、RTX 4090等消费级显卡即可本地运行。\n沐曦此次Day 0适配的核心支撑来自MXMACA软件栈。该软件栈全链路覆盖底层驱动、MXCC编译器、算子深度适配及主流训练/推理框架对接，可将传统模型适配周期从数周压缩至小时级。\n目前MXMACA已兼容40余种AI框架，适配超1000款模型，全量支持PyTorch 2.8的2410个GPU算子。\n自2025年12月以来，沐曦已完成38个主流顶尖模型的Day 0适配，范围涵盖智谱AI、阿里千问、MiniMax、DeepSeek、阶跃星辰、腾讯混元等头部厂商，适配速度、数量与生态广度均位居行业前列。","source":"新浪财经_金融信息服务商","date":"2026-09-19T13:26:03+00:00","content":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配\n（来源：快科技）\n快科技9月19日消息，中电信人工智能科技有限公司日前正式发布新一代星辰大模型Xing4.0-29B-A4B。沐曦股份宣布，依托自研MXMACA软件栈，曦云C系列GPU率先完成对该模型的Day 0适配，实现\"上线即适配\"。\nXing4.0-29B-A4B是面向智能体时代打造的轻量级代码智能体大模型，采用MOE架构，总参数量29B，激活参数仅4B，原生支持256K上下文，可扩展至512K。\n该模型是国内首个基于国产算力与国产框架完成训练、面向复杂工程任务深度优化的百亿参数大模型，聚焦任务规划、工具调用和自主执行等能力。\n性能方面，Xing4.0-29B-A4B在SWE-bench Verified上得分75.0，Terminal-Bench 2.1得分57.5，均超过同尺寸竞品。\n在SuperCLUE智能体能力评测中，该模型以93.52分位列第三，与两款头部Qwen模型差距不足1分。4-bit量化后显存占用降至15GB，RTX 3090、RTX 4090等消费级显卡即可本地运行。\n沐曦此次Day 0适配的核心支撑来自MXMACA软件栈。该软件栈全链路覆盖底层驱动、MXCC编译器、算子深度适配及主流训练/推理框架对接，可将传统模型适配周期从数周压缩至小时级。\n目前MXMACA已兼容40余种AI框架，适配超1000款模型，全量支持PyTorch 2.8的2410个GPU算子。\n自2025年12月以来，沐曦已完成38个主流顶尖模型的Day 0适配，范围涵盖智谱AI、阿里千问、MiniMax、DeepSeek、阶跃星辰、腾讯混元等头部厂商，适配速度、数量与生态广度均位居行业前列。","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//finance.sina.cn/stock/jdts/2026-09-19/detail-iniskptr2203813.d.html%3Foid%3D800%26vt%3D4%26cid%3D76993%26node_id%3D76993","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 749 characters.","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 749 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":749,"summary_length":749,"usable_text_length":749,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":749,"summary_length":749}},"tags":[]},"fallback_formats":["markdown","json","html"],"actions":{"read":"/item/85865","export_markdown":"/api/items/85865/export?format=markdown","export_json":"/api/items/85865/export?format=json","diagnose":"/api/diagnose?url=https%3A//finance.sina.cn/stock/jdts/2026-09-19/detail-iniskptr2203813.d.html%3Foid%3D800%26vt%3D4%26cid%3D76993%26node_id%3D76993"},"formats":{"full":{"id":85865,"title":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配 - 新浪财经_金融信息服务商","url":"https://finance.sina.cn/stock/jdts/2026-09-19/detail-iniskptr2203813.d.html?oid=800&vt=4&cid=76993&node_id=76993","source":"新浪财经_金融信息服务商","author":null,"published_at":"2026-09-19T13:26:03+00:00","locale":"zh","topic":"ai","tags":[],"excerpt":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配\n（来源：快科技）\n快科技9月19日消息，中电信人工智能科技有限公司日前正式发布新一代星辰大模型Xing4.0-29B-A4B。沐曦股份宣布，依托自研MXMACA软件栈，曦云C系列GPU率先完成对该模型的Day 0适配，实现\"上线即适配\"。\nXing4.0-29B-A4B是面向智能体时代打造的轻量级代码智能体大模型，采用MOE架构，总参数量29B，激活参数仅4B，原生支持256K上下文，可扩展至512K。\n该模型是国内首个基于国产算力与国产框架完成训练、面向复杂工程任务深度优化的百亿参数大模型，聚焦任务规划、工具调用和自主执行等能力。\n性能方面，Xing4.0-29B-A4B在SWE-bench Verified上得分75.0，Terminal-Bench 2.1得分57.5，均超过同尺寸竞品。\n在SuperCLUE智能体能力评测中，该模型以93.52分位列第三，与两款头部Qwen模型差距不足1分。4-bit量化后显存占用降至15GB，RTX 3090、RTX 4090等消费级显卡即可本地运行。\n沐曦此次Day 0适配的核心支撑来自MXMACA软件栈。该软件栈全链路覆盖底层驱动、MXCC编译器、算子深度适配及主流训练/推理框架对接，可将传统模型适配周期从数周压缩至小时级。\n目前MXMACA已兼容40余种AI框架，适配超1000款模型，全量支持PyTorch 2.8的2410个GPU算子。\n自2025年12月以来，沐曦已完成38个主流顶尖模型的Day 0适配，范围涵盖智谱AI、阿里千问、MiniMax、DeepSeek、阶跃星辰、腾讯混元等头部厂商，适配速度、数量与生态广度均位居行业前列。","full_text":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配\n（来源：快科技）\n快科技9月19日消息，中电信人工智能科技有限公司日前正式发布新一代星辰大模型Xing4.0-29B-A4B。沐曦股份宣布，依托自研MXMACA软件栈，曦云C系列GPU率先完成对该模型的Day 0适配，实现\"上线即适配\"。\nXing4.0-29B-A4B是面向智能体时代打造的轻量级代码智能体大模型，采用MOE架构，总参数量29B，激活参数仅4B，原生支持256K上下文，可扩展至512K。\n该模型是国内首个基于国产算力与国产框架完成训练、面向复杂工程任务深度优化的百亿参数大模型，聚焦任务规划、工具调用和自主执行等能力。\n性能方面，Xing4.0-29B-A4B在SWE-bench Verified上得分75.0，Terminal-Bench 2.1得分57.5，均超过同尺寸竞品。\n在SuperCLUE智能体能力评测中，该模型以93.52分位列第三，与两款头部Qwen模型差距不足1分。4-bit量化后显存占用降至15GB，RTX 3090、RTX 4090等消费级显卡即可本地运行。\n沐曦此次Day 0适配的核心支撑来自MXMACA软件栈。该软件栈全链路覆盖底层驱动、MXCC编译器、算子深度适配及主流训练/推理框架对接，可将传统模型适配周期从数周压缩至小时级。\n目前MXMACA已兼容40余种AI框架，适配超1000款模型，全量支持PyTorch 2.8的2410个GPU算子。\n自2025年12月以来，沐曦已完成38个主流顶尖模型的Day 0适配，范围涵盖智谱AI、阿里千问、MiniMax、DeepSeek、阶跃星辰、腾讯混元等头部厂商，适配速度、数量与生态广度均位居行业前列。","reading_time_min":1,"extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 749 characters.","diagnostics_url":"/api/diagnose?url=https%3A//finance.sina.cn/stock/jdts/2026-09-19/detail-iniskptr2203813.d.html%3Foid%3D800%26vt%3D4%26cid%3D76993%26node_id%3D76993","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 749 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":749,"summary_length":749,"usable_text_length":749,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":749,"summary_length":749}}},"quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 749 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":749,"summary_length":749,"usable_text_length":749,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":749,"summary_length":749}},"actions":{"read":"/item/85865","export_markdown":"/api/items/85865/export?format=markdown","export_json":"/api/items/85865/export?format=json","diagnose":"/api/diagnose?url=https%3A//finance.sina.cn/stock/jdts/2026-09-19/detail-iniskptr2203813.d.html%3Foid%3D800%26vt%3D4%26cid%3D76993%26node_id%3D76993"}},"digest":{"id":85865,"title":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配 - 新浪财经_金融信息服务商","url":"https://finance.sina.cn/stock/jdts/2026-09-19/detail-iniskptr2203813.d.html?oid=800&vt=4&cid=76993&node_id=76993","source":"新浪财经_金融信息服务商","topic":"ai","published_at":"2026-09-19T13:26:03+00:00","excerpt":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配 （来源：快科技） 快科技9月19日消息，中电信人工智能科技有限公司日前正式发布新一代星辰大模型Xing4.0-29B-A4B。沐曦股份宣布，依托自研MXMACA软件栈，曦云C系列GPU率先完成对该模型的Day 0适配，实现\"上线即适配\"。 Xing4.0-29B-A4B是面向智能体时代打造的轻量级代码智能体大模型，采用MOE架构，总参数量29B，激活参数仅4B，原生支持256K上下文，可扩展至512K。…","quality_bucket":"high","quality_reason":"High confidence: full text extraction produced 749 characters.","reading_time_min":1,"cluster_id":3513420},"card":{"display_title":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配 - 新浪财经_金融信息服务商","subtitle":"新浪财经_金融信息服务商 · 2026-09-19","summary":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配 （来源：快科技） 快科技9月19日消息，中电信人工智能科技有限公司日前正式发布新一代星辰大模型Xing4.0-29B-A4B。沐曦股份宣布，依托自研MXMACA软件栈，曦云C系列GPU率先完成对该模型的Day 0适配，实现\"上线即适配\"。…","badges":["quality:high"],"links":{"read":"/item/85865","original":"https://finance.sina.cn/stock/jdts/2026-09-19/detail-iniskptr2203813.d.html?oid=800&vt=4&cid=76993&node_id=76993","diagnose":"/api/diagnose?url=https%3A//finance.sina.cn/stock/jdts/2026-09-19/detail-iniskptr2203813.d.html%3Foid%3D800%26vt%3D4%26cid%3D76993%26node_id%3D76993"},"quality_warning":null},"export":{"title":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配 - 新浪财经_金融信息服务商","url":"https://finance.sina.cn/stock/jdts/2026-09-19/detail-iniskptr2203813.d.html?oid=800&vt=4&cid=76993&node_id=76993","summary":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配\n（来源：快科技）\n快科技9月19日消息，中电信人工智能科技有限公司日前正式发布新一代星辰大模型Xing4.0-29B-A4B。沐曦股份宣布，依托自研MXMACA软件栈，曦云C系列GPU率先完成对该模型的Day 0适配，实现\"上线即适配\"。\nXing4.0-29B-A4B是面向智能体时代打造的轻量级代码智能体大模型，采用MOE架构，总参数量29B，激活参数仅4B，原生支持256K上下文，可扩展至512K。\n该模型是国内首个基于国产算力与国产框架完成训练、面向复杂工程任务深度优化的百亿参数大模型，聚焦任务规划、工具调用和自主执行等能力。\n性能方面，Xing4.0-29B-A4B在SWE-bench Verified上得分75.0，Terminal-Bench 2.1得分57.5，均超过同尺寸竞品。\n在SuperCLUE智能体能力评测中，该模型以93.52分位列第三，与两款头部Qwen模型差距不足1分。4-bit量化后显存占用降至15GB，RTX 3090、RTX 4090等消费级显卡即可本地运行。\n沐曦此次Day 0适配的核心支撑来自MXMACA软件栈。该软件栈全链路覆盖底层驱动、MXCC编译器、算子深度适配及主流训练/推理框架对接，可将传统模型适配周期从数周压缩至小时级。\n目前MXMACA已兼容40余种AI框架，适配超1000款模型，全量支持PyTorch 2.8的2410个GPU算子。\n自2025年12月以来，沐曦已完成38个主流顶尖模型的Day 0适配，范围涵盖智谱AI、阿里千问、MiniMax、DeepSeek、阶跃星辰、腾讯混元等头部厂商，适配速度、数量与生态广度均位居行业前列。","source":"新浪财经_金融信息服务商","date":"2026-09-19T13:26:03+00:00","content":"国产算力新里程碑！曦云C系列GPU率先完成中国电信星辰Xing4.0大模型Day 0适配\n（来源：快科技）\n快科技9月19日消息，中电信人工智能科技有限公司日前正式发布新一代星辰大模型Xing4.0-29B-A4B。沐曦股份宣布，依托自研MXMACA软件栈，曦云C系列GPU率先完成对该模型的Day 0适配，实现\"上线即适配\"。\nXing4.0-29B-A4B是面向智能体时代打造的轻量级代码智能体大模型，采用MOE架构，总参数量29B，激活参数仅4B，原生支持256K上下文，可扩展至512K。\n该模型是国内首个基于国产算力与国产框架完成训练、面向复杂工程任务深度优化的百亿参数大模型，聚焦任务规划、工具调用和自主执行等能力。\n性能方面，Xing4.0-29B-A4B在SWE-bench Verified上得分75.0，Terminal-Bench 2.1得分57.5，均超过同尺寸竞品。\n在SuperCLUE智能体能力评测中，该模型以93.52分位列第三，与两款头部Qwen模型差距不足1分。4-bit量化后显存占用降至15GB，RTX 3090、RTX 4090等消费级显卡即可本地运行。\n沐曦此次Day 0适配的核心支撑来自MXMACA软件栈。该软件栈全链路覆盖底层驱动、MXCC编译器、算子深度适配及主流训练/推理框架对接，可将传统模型适配周期从数周压缩至小时级。\n目前MXMACA已兼容40余种AI框架，适配超1000款模型，全量支持PyTorch 2.8的2410个GPU算子。\n自2025年12月以来，沐曦已完成38个主流顶尖模型的Day 0适配，范围涵盖智谱AI、阿里千问、MiniMax、DeepSeek、阶跃星辰、腾讯混元等头部厂商，适配速度、数量与生态广度均位居行业前列。","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//finance.sina.cn/stock/jdts/2026-09-19/detail-iniskptr2203813.d.html%3Foid%3D800%26vt%3D4%26cid%3D76993%26node_id%3D76993","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 749 characters.","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 749 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":749,"summary_length":749,"usable_text_length":749,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":749,"summary_length":749}},"tags":[],"format_contract_version":"news_item_formats.v1"}}}