SpaceXAI Releases Grok 4.7: A Larger Base Model at the Same $2/$6 Price as Grok 4.6
https://www.marktechpost.com/2026/09/21/spacexai-releases-grok-4-7/📌 Grok 4.7 同價登場，Terminal-Bench 分數翻近一倍
TL;DR：SpaceXAI 用更大 base model 與更長 RL 訓練推出 Grok 4.7，維持與 4.6 相同的 2 美元/6 美元定價，多項 benchmark 明顯進步但非全面領先。
同樣的價格，換來大幅提升的 coding 與 agentic 表現，SpaceXAI 這次打的是「免費升級」牌。Grok 4.7 已經可以透過 xAI API、Cursor、Grok Build、OpenRouter、Vercel、Cloudflare 呼叫使用。
🤔 更大的底子，更長的訓練
Grok 4.7 是 SpaceXAI 針對 coding、agentic 任務與知識工作推出的新旗艦模型，建立在更大的 base model 與更長的強化學習（RL）訓練之上，但售價與速度都維持在 Grok 4.6 的水準。
📊 跑分全面優於 4.6，但不是每項都拿第一
發布時的比較表格把 Grok 4.7（xHigh effort）拿去對比 Grok 4.6（High）、GPT-5.6 Sol Max、Fable 5.1 Max（所有分數皆為廠商自報）：
| Benchmark | Grok 4.7 | Grok 4.6 | 備註 |
|---|---|---|---|
| Terminal-Bench 4.0 | 38.0% | 20.3% | 表中最大進步幅度 |
| EEBench | 64.0% | 53.0% | 表中最高分 |
| Harvey 法務 agent benchmark | 19.6% | — | 對比 Fable 5.1 Max 的 6.7% |
| GDPval (xhigh/high Elo) | 1695 | 1605 | 專業知識工作測試 |
Grok 4.7 在每一項對比 Grok 4.6 的指標上都有進步。不過放到更大的競爭格局裡，它並非全面領先：Fable 5.1 Max 在 7 項 benchmark 中拿下 4 項第一，包括 57.9% 的 Terminal-Bench 分數與 1735 的 GDPval Elo；GPT-5.6 Sol Max 則在 DeepSWE v1.1 上以 72.7% 保持領先（Grok 4.7 的 DeepSWE 分數是以 high effort 跑出）。GPT-6 Astra max 的 GDPval 分數為 1542。
💡 價格才是 Grok 4.7 真正的賣點
比較貴的對手貴不少：Fable 5.1 Max 的輸入價格是 Grok 4.7 的 5 倍，輸出約 8.3 倍；GPT-5.6 Sol Max 輸入貴 2 倍，輸出約貴 3.3 倍。SpaceXAI 在 CursorBench 4.0 的成本對任務效益圖表中，把 Grok 4.7 定位在價效比前緣。SpaceXAI 也表示 Grok 4.7 在製作文件與簡報上有所提升。
🧩 安全性與部署細節
Grok 4.7 搭配全新的安全防護堆疊，SpaceXAI 稱其為目前測試過在拒答與 jailbreak 抵抗力上最強的版本，在 LatchBio 的生物安全 benchmark 拿下 62.4% 的最高分；在 SpaceXAI 自家的 HackerBench v0.3（針對高風險、雙重用途網路安全任務）測試中，只放行了 3.3% 的風險 prompt，官方表示這幾乎不會擋到合法的資安研究工作，並開放特定資安合作夥伴以邀請制方式使用其紅隊能力做防禦研究。
定價維持 $2/百萬輸入 token、$6/百萬輸出 token，Cursor 全方案可用，並是 Grok Build 的預設模型，也開放透過 Grok API、OpenRouter、Vercel、Cloudflare 使用。另有 Grok 4.7 Fast 版本，跑在更快的基礎設施上，輸出速度快兩倍、價格也貴兩倍，但只能在 Cursor 與 Grok Build 使用，不開放公開 API，也不含在 Grok Build 免費方案內。想把推論留在美國境內的話，https://us.api.x.ai/v1 這個區域端點會加收 10% 費用。官方也建議設定 prompt_cache_key 以確保 cache 命中率。
⚠️ 沒有全面領先，適合誰要看場景
Grok 4.7 的優勢集中在對 Grok 4.6 的進步幅度與價效比，但在 DeepSWE、GDPval 等指標上，GPT-5.6 Sol Max 與 Fable 5.1 Max 仍分別保有領先，選型時仍需按實際任務對照對應 benchmark。
🎯 實務啟示
如果團隊原本就在用 Grok 4.6，這次升級不需要重新評估預算，$2/$6 的定價不變，卻在 Terminal-Bench、EEBench 等 agentic 相關指標上有明顯進步，值得直接換上；若在意的是 DeepSWE 或頂級 GDPval 分數，GPT-5.6 Sol Max、Fable 5.1 Max 仍是對照組，需要按價格與分數一起權衡。
🔗 來源
- 標題：SpaceXAI Releases Grok 4.7: A Larger Base Model at the Same $2/$6 Price as Grok 4.6
- 作者／機構：Michal Sutter @ MarkTechPost
- 連結：https://www.marktechpost.com/2026/09/21/spacexai-releases-grok-4-7/
#Grok47 #SpaceXAI #LLM #CodingAgent #AIBenchmark #TerminalBench #AgenticAI #AISafety #ModelRelease #CursorBench
原始資料 MarkTechPost · 收集於 2026-09-22
摘要原文
SpaceXAI has released Grok 4.7 , its new flagship model for coding, agentic tasks, and knowledge work. Grok 4.7 is built on a larger base model and a longer reinforcement learning run. It still ships at the same price and speed as Grok 4.6 . Is it deployable? Yes, as a hosted model. You can call grok-4.7 today through the xAI API, Cursor, Grok Build, OpenRouter, Vercel, and Cloudflare. SpaceXAI lists 4 changes over Grok 4.6: The developer docs list the API specs: The launch table compares Grok 4.7 at xHigh effort with Grok 4.6 High, GPT-5.6 Sol Max, and Fable 5.1 Max. The Grok 4.7 DeepSWE score was run at high effort. All scores are vendor-reported. Grok 4.7 improves on Grok 4.6 in every row. The largest jump is on Terminal-Bench 4.0, from 20.3% to 38.0%. EEBench rose 11 points to 64.0%, the top score in the table. On Harvey’s legal agent benchmark, Grok 4.7 scored 19.6% against 6.7% for Fable 5.1 Max. Grok 4.7 does not lead across the board. Fable 5.1 Max tops 4 of 7 benchmarks, including a 57.9% Terminal-Bench score. GPT-5.6 Sol Max holds the top DeepSWE v1.1 result at 72.7%. Price is the other axis. Fable 5.1 Max costs 5x more on input and about 8.3x more on output. GPT-5.6 Sol Max costs 2x more on input and about 3.3x more on output. On a CursorBench 4.0 cost-per-task chart, SpaceXAI places Grok 4.7 at the frontier in price-performance. On GDPval, which tests professional knowledge work, Grok 4.7 xhigh scored 1,695 Elo. That is up from 1,605 for Grok 4.6 high. Fable 5.1 max leads at 1,735, and GPT-6 Astra max scored 1,542. SpaceXAI also says Grok 4.7 is better at creating documents and presentations. Grok 4.7 ships with an entirely new safeguard stack. SpaceXAI calls it the strongest model it has tested on refusals and jailbreak resistance. It topped LatchBio’s biosafety benchmark at 62.4%. On HackerBench v0.3, SpaceXAI’s own benchmark for risky and malicious cyber tasks, the model let 3.3% of risky dual-use prompts through. The company says it rarely blocks legitimate security work. Select cybersecurity partners now get invite-only access to its red-team capabilities for defense research. Grok 4.7 costs $2 per million input tokens and $6 per million output tokens. It is available in Cursor on all plans and is the default model in Grok Build . It is also served through the Grok API , OpenRouter, Vercel, and Cloudflare. Grok 4.7 Fast is the same model on faster infrastructure, with twice the output speed at twice the price. The docs say it runs only in Cursor and Grok Build, not on the public xAI API. It is also excluded from Grok Build’s free tier. A US regional endpoint at https://us.api.x.ai/v1 keeps inference in the United States at a 10% premium. SpaceXAI recommends setting a prompt_cache_key for reliable cache hits. Check out the announcement , docs , and X post . All credit goes to the researcher of this project. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter . Wait! are you on telegram? now you can join us on telegram as well. Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us The post SpaceXAI Releases Grok 4.7: A Larger Base Model at the Same $2/$6 Price as Grok 4.6 appeared first on MarkTechPost .
由 claude-code/sonnet 自動生成