[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-grok-4-5-1-5t-v9-cursor-beta":3,"news-related-bce04851-f4a1-410c-b543-d08e45eb4a37":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"bce04851-f4a1-410c-b543-d08e45eb4a37","Grok 4.5 私测启动：1.5T V9 + Cursor 数据，宣称对标 Opus 4.6","马斯克 6 月 28 日通过 X 宣布，xAI 最新大语言模型 **Grok 4.5** 已在 **SpaceX 和特斯拉内部开启 Beta 测试**——这是 Grok 系列首次把内测放在企业级「真生产环境」评估。\n\n## 模型本体\n\n- 底座为 **V9 基础模型，1.5 万亿参数**，比 5 月底 Grok 4.4（1T）扩张 50%\n- 「补充训练」混入 **Cursor 开发者工作流数据**，把 IDE 真实使用痕迹灌进对齐\n- **RL 仍在持续优化**，马斯克称「每一轮 RL 都显著提升能力」\n\n## 性能对标\n\n内部早期评估显示 **「接近、可能超过 Claude Opus」**（指向 Opus 4.6），是 Grok 首次公开对标前沿闭源模型。但**目前无独立外部 benchmark**，结论仍属厂内口径。\n\n## 节奏重塑\n\n马斯克同步宣布：**SpaceX 将在 2026 余下时间内每月发布一款「完全从零训练」的全新基础模型**——同行大版本节奏仍为 3–6 个月。这是 xAI 把 Colossus 集群算力直接转译为「交付节奏」的最激进表态。\n\n## 点评\n\n把 1.5T 模型塞进 SpaceX\u002FTesla 内部做 beta，本质是用两个世界级工程团队做大规模 agent 评估。**Cursor 数据引入值得注意**：当 IDE 工作流进入训练，模型就从「会写代码」变成「会像 Cursor 用户一样写代码」——SaaS 工具厂商从「应用层」被拉到了「数据供应商」层。月度从零训练的承诺能否兑现，要看 Colossus 产能和训练流水线成熟度。","https:\u002F\u002F36kr.com\u002Fnewsflashes\u002F3873872147272963","b82e17a3-1dbd-4b5d-88dc-9f518f917cc0",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":18,"name":19,"slug":19,"description":13,"color":13},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":21,"name":22,"slug":22,"description":13,"color":13},"b1853a5a-d940-42b7-94f9-0488ee3f2cf7","new-model",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"46d3d788-0b73-4425-8e97-506e7234ba0a","en","Grok 4.5 private beta: 1.5T V9 trained on Cursor data","On June 28, Elon Musk announced via X that xAI's newest large language model, Grok 4.5, has begun a beta test inside SpaceX and Tesla — the first time the Grok series has run an internal test in a true enterprise-grade production environment.\n\nThe model itself: built on a V9 base of 1.5 trillion parameters, 50% larger than the 1T Grok 4.4 from late May; supplementary training mixes in Cursor developer-workflow data, baking real IDE usage into alignment; RL is still being optimized, and Musk claims every round of RL significantly improves capability.\n\nInternal early evaluation shows \"close to, possibly exceeding Claude Opus\" (pointing at Opus 4.6) — the first time Grok has publicly benchmarked itself against a frontier closed model. But there are still no independent external benchmarks, and the conclusion remains a vendor-side claim.\n\nMusk also announced: SpaceX will release a brand-new foundation model, trained completely from scratch, every month for the rest of 2026 — versus the 3-6 month major-release cadence of peers. This is xAI's most aggressive translation of Colossus-cluster compute directly into delivery cadence.\n\nCommentary: putting a 1.5T model in production beta at SpaceX and Tesla is, in effect, running a massive agent evaluation with two world-class engineering teams. The Cursor-data inclusion is notable: once IDE workflow enters training, the model goes from \"knows how to write code\" to \"writes code the way a Cursor user does\" — SaaS-tool vendors are pulled from the \"application layer\" into the \"data provider\" layer. Whether the monthly from-scratch-training promise holds depends on Colossus capacity and training-pipeline maturity.","grok-4-5-1-5t-v9-cursor-beta","2026-06-29T10:01:00Z","2026-06-29T10:08:14.603529Z","2026-08-19T02:08:40.142862Z",true,"agent",116,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"f9cf9f03-6aca-4d29-94d3-5c6acfeaf435","匿名模型 OX Alpha 短暂登顶 OpenRouter 编码榜:研究者推测底座指向智谱 GLM-5.x","ox-alpha-stealth-openrouter-glm-5-zhipu","2026-08-24T03:00:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"dccdd14b-babe-4883-b0ec-0cf5b1d85018","MiniMax Music 3.0 把「5 分钟完整歌曲」开源:Hybrid-LM + Flow-VAE 让音乐生成跨过录音室门槛","minimax-music-3-5min-song-open-source","2026-08-15T00:00:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"49d19ba1-8f45-475c-bed1-a69dc353523e","字节跳动用 10 万亿参数下注：规模赛跑与张一鸣的「不蒸馏」表态","bytedance-10t-mythos-zhangyiming-no-distill-2026-08","2026-08-08T00:00:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"70b5b0d6-ce28-48e8-abe4-6a667a723c4e","xAI 把 Colossus 推到 2 GW:555,000 颗 GPU 撑起 Grok 4.6\u002F4.7 的万亿参数竞速","xai-colossus-2gw-grok-4-6-7-compute","2026-07-31T04:00:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"f4af1e4e-98a3-4810-831c-699ffe31ae73","马斯克公布 Grok 4.6\u002F4.7 路线图：1.5T\u002F2.1T 参数，SFT+RL 升级，8 月 7 日发行","grok-4-6-4-7-roadmap-1-5t-2-1t","2026-07-30T08:45:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"dea38861-2618-4468-9bad-a18eea96a818","Base44 Base 1：年入 1.5 亿美元的 vibe-coding 平台，终于把自己的 LLM 训出来","base44-base-1-vibe-coding-llm-launch","2026-07-29T06:00:00+00:00"]