SpaceXAI发布Grok 4.5：1.5万亿参数，首个编程智能体模型对标Claude Opus | SpaceXAI Launches Grok 4.5: 1.5T Params, First Coding Agent Model Taking Aim at Claude Opus
SpaceXAI发布Grok 4.5：1.5万亿参数，首个编程智能体模型对标Claude Opus
7月9日，马斯克旗下SpaceXAI正式向公众开放Grok 4.5。这是SpaceXAI更名后发布的首个专门针对编程和智能体任务训练的模型，参数规模约1.5万亿，约为上一代的3倍。与GPT-5.6同日发布，7月9日成为大模型竞争最密集的一天。
从xAI到SpaceXAI：品牌背后的战略转向
SpaceXAI这个名字本身就传递了马斯克的野心——将AI能力与SpaceX的航天工程深度绑定。Grok 4.5是更名后的首个旗舰发布，选择编程和智能体作为主攻方向，显然是冲着Anthropic的Claude Code和OpenAI的Codex去的。
1.5万亿参数意味着什么
Grok 4.5基于V9基础架构，参数规模约1.5万亿。作为对比：
- GPT-5.5参数约1万亿（估算）
- Claude Opus 4.8参数未公开，业界估计在1-2万亿之间
- Grok 3参数约5000亿
3倍的参数跳跃并不常见。通常模型迭代会在架构优化和训练数据质量上下功夫，而非单纯堆参数。但马斯克选择了暴力美学路线——更大的模型、更多的数据、更强的算力。
“Opus级别”的定位
马斯克将Grok 4.5描述为”Opus级别”——对标Anthropic Claude Opus，但声称速度更快、token使用效率更高、成本更低。
这三个维度的对比很有意思：
- 速度：Grok 4.5的推理速度确实是一个卖点，SpaceXAI自研的推理优化技术可能在这里发挥了作用
- Token效率：如果同样任务消耗更少的token，对开发者来说是实打实的成本节省
- 成本：低价是抢占市场的利器，但也意味着SpaceXAI在用利润换份额
编程智能体：新赛道
Grok 4.5被定位为”首个编程智能体模型”——这不是简单的代码补全工具，而是能够自主完成多步骤编程任务的Agent。
此前SpaceXAI已与Cursor达成联合模型合作。Grok 4.5在编程智能体场景的表现将成为市场关注的焦点。如果Grok 4.5能在Cursor等编辑器中表现出色，它将直接挑战Claude Code在企业编程市场的统治地位。
同日发布的竞争意义
GPT-5.6与Grok 4.5同日发布是一个巧合，但也是一个信号：大模型发布正在从”各自为战”转向”正面交锋”。当多家公司在同一天发布旗舰产品，市场的注意力被分散，每家都希望在同一天的新闻洪流中脱颖而出。
对开发者来说，这其实是好事——更多的选择、更低的价格、更强的能力。但对模型公司来说，利润空间正在被压缩。
下周更热闹
7月15日DeepSeek V4上线，7月17日谷歌Gemini 3.5 Pro发布。三款旗舰模型（GPT-5.6、DeepSeek V4、Gemini 3.5 Pro）将在同一周内正面交锋。加上今天的Grok 4.5，7月将成为2026年大模型竞争最激烈的月份。
SpaceXAI Launches Grok 4.5: 1.5T Params, First Coding Agent Model Taking Aim at Claude Opus
On July 9, Elon Musk’s SpaceXAI officially opened Grok 4.5 to the public. This is the first model trained specifically for coding and agent tasks after the SpaceXAI rebrand, with approximately 1.5 trillion parameters — about 3x the previous generation. Released on the same day as GPT-5.6, July 9 became one of the most competitive days in the large model race.
From xAI to SpaceXAI: The Strategy Behind the Brand
The SpaceXAI name itself conveys Musk’s ambition — deeply integrating AI capabilities with SpaceX’s aerospace engineering. Grok 4.5 is the first flagship release after the rebrand, targeting coding and agents, clearly aimed at Anthropic’s Claude Code and OpenAI’s Codex.
What 1.5 Trillion Parameters Means
Grok 4.5 is built on the V9 architecture with approximately 1.5 trillion parameters. For comparison: GPT-5.5 is estimated at ~1T, Claude Opus 4.8 estimated at 1-2T, and Grok 3 at ~500B. A 3x parameter jump is uncommon. Most model iterations focus on architecture optimization and training data quality rather than simply scaling parameters. But Musk chose brute force — bigger models, more data, more compute.
“Opus-Level” Positioning
Musk described Grok 4.5 as “Opus-level” — benchmarked against Claude Opus, but claiming faster speed, better token efficiency, and lower cost. These three dimensions are interesting: speed may come from SpaceXAI’s proprietary inference optimization; token efficiency means real cost savings for developers; low pricing is a market-share grabber but means SpaceXAI is trading profit for share.
Coding Agent: A New Track
Grok 4.5 is positioned as the “first coding agent model” — not just a code completion tool, but an Agent capable of autonomously completing multi-step programming tasks. SpaceXAI previously partnered with Cursor on a joint model. Grok 4.5’s performance in coding agent scenarios will be a market focus. If it performs well in editors like Cursor, it will directly challenge Claude Code’s dominance in the enterprise coding market.
Next Week Gets Hotter
DeepSeek V4 launches July 15, Google Gemini 3.5 Pro on July 17. Three flagship models (GPT-5.6, DeepSeek V4, Gemini 3.5 Pro) will clash in the same week. Combined with today’s Grok 4.5, July will be the most competitive month for large models in 2026.