[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-grok-4-5-public-launch":3,"news-related-d0d290cd-c169-495c-b0be-b559caa6eaa0":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},"d0d290cd-c169-495c-b0be-b559caa6eaa0","Grok 4.5 公众开放定档 7月9日:Musk 押注的不是参数,而是「Opus 级但更便宜」","7月8日,马斯克在 X 宣布 SpaceXAI 将于 7月9日正式向公众开放 Grok 4.5——这是该模型自6月28日进入 SpaceX 与 Tesla 内部 beta 后的首次公开亮相,也是 SpaceX 完成对 Cursor 收购后的第一款\"血统清晰\"的产出。Grok 4.5 基于 1.5T 参数的 V9 底座,补入 Cursor IDE 的真实交互数据做增量训练:不是静态代码语料,而是多轮编辑、错误恢复、跨大仓库上下文这类 agentic coding 信号。The Information 因此把它定位为 SpaceXAI 与 Cursor 的\"首个联合模型\"——模型实验室 + IDE 数据 + Memphis Colossus 的 22 万张 GPU,现在属于同一家。马斯克这次的卖点不是参数,而是\"Opus 级、更快、更省 token、更便宜\"。**更便宜才是真正的钩子**:开发者钱包的战场正从\"谁的 benchmark 高\"转到\"每任务多少钱\"。同一周,Anthropic 因 Claude Opus 4.7 的 tokenizer 调价被批评,OpenAI 正把 GPT-5.6 Sol 推上 7月9日的 GA,Musk 把价格维度提到台面,是在抢 OpenAI\u002FAnthropic 的企业付费心智。但有一件事没解决:Musk 没有公布 SWE-Bench、Terminal-Bench、GPQA 等任何公开跑分。\"Opus 级\"在 4.5\u002F4.7\u002F4.8 之间差距显著,**没有数字就只能等到 7月9日**。如果 Grok 4.5 真能贴近 Opus 4.8 的 69.2% SWE-Bench Pro 水平,加上 token 效率,确实可能撬动一部分 Cursor 用户——但\"Opus 级\"必须从口号变成可采购的规格才算数。7月9日是 2026 年目前最拥挤的\"前沿模型日\":Grok 4.5 与 GPT-5.6 Sol 同日 GA,Anthropic 的 Fable 5 在解封后持续放量。**Grok 4.5 的真正考验,是 24 小时后的第一批独立跑分**——单选一家的日子已经结束,接下来拼的是路由与成本工程。","https:\u002F\u002F36kr.com\u002Fnewsflashes\u002F3886610502463491","5e4fd3d1-9cb4-44a6-bae5-9ffb449c05c1",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",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},"12399331-4e5e-438a-afba-33024a648bba","en","Grok 4.5 opens July 9: Musk bets on Opus-class but cheaper","On July 8, Musk announced on X that SpaceXAI will officially open Grok 4.5 to the public on July 9 — the model's first public appearance after entering SpaceX and Tesla internal beta on June 28, and also the first \"clean bloodline\" product after SpaceX's acquisition of Cursor. Grok 4.5 is based on a 1.5T-parameter V9 base, supplemented with Cursor IDE's real interaction data for incremental training: not static code corpus, but multi-round editing, error recovery, cross-large-repository context and other agentic coding signals. The Information therefore positions it as the \"first joint model\" between SpaceXAI and Cursor — model lab + IDE data + Memphis Colossus's 220,000 GPUs, now under the same roof. Musk's selling point this time isn't parameters, but \"Opus grade, faster, more token-efficient, cheaper\". **Cheaper is the real hook**: the developer's wallet battlefield is shifting from \"whose benchmark is higher\" to \"how much per task\". The same week, Anthropic was criticized for Claude Opus 4.7's tokenizer repricing, OpenAI is pushing GPT-5.6 Sol to GA on July 9, and Musk is putting the price dimension on the table, going for the OpenAI\u002FAnthropic enterprise paid mindshare. But one thing isn't resolved: Musk hasn't published any public scores such as SWE-Bench, Terminal-Bench, GPQA. \"Opus grade\" differs significantly between 4.5\u002F4.7\u002F4.8, **without numbers you'll just have to wait until July 9**. If Grok 4.5 can really get close to Opus 4.8's 69.2% SWE-Bench Pro level, plus token efficiency, it could indeed pry some Cursor users away — but \"Opus grade\" has to go from slogan to a buyable spec. July 9 is the most crowded \"frontier model day\" of 2026 so far: Grok 4.5 and GPT-5.6 Sol GA on the same day, Anthropic's Fable 5 continues to scale after being unfrozen. **Grok 4.5's real test is the first batch of independent scores 24 hours later** — the single-vendor era is over, what comes next is a contest of routing and cost engineering.","grok-4-5-public-launch","2026-07-08T12:30:00Z","2026-07-08T12:05:54.668611Z","2026-08-19T02:08:40.142862Z",true,"agent",100,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"97a4f24e-c65b-4e95-b9ed-b05997d88af2","Qwen3.6-35B-A3B发布：开源大模型的效率革命","qwen-3-6-35b-a3b-open-source-coding-262k","2026-04-25T02:05:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"5878a668-282c-4b88-b2b8-7eef40b7938c","LFM2.5-2.6B：2.5GB 内存跑本机 Agent 220 tok\u002Fs","lfm2-5-2-6b-on-device-agent","2026-08-11T00:00:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"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":54,"title":55,"news_slug":56,"published_at":57},"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":59,"title":60,"news_slug":61,"published_at":62},"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",{"id":64,"title":65,"news_slug":66,"published_at":67},"1ecbb79a-d843-43ad-b533-c01ae396275f","Qwen-Audio-3.0-Realtime：蒸馏拉满实时语音智商与延迟","qwen-audio-3-realtime","2026-07-15T10:00:00+00:00"]