xAI releases Grok 4.7 with improved self-verification, longer context, at $2/$6 per 1M tokens
xAI released Grok 4.7, a larger base model with extended reinforcement learning, claiming twice the speed and half the cost of comparable models for coding and knowledge work. The model shows gains on long-horizon benchmarks like CursorBench 4.0 (46.3% vs 40.4% for Grok 4.6) and Terminal-Bench 4.0 (38.0% vs 20.3%), though some observers note benchmark scores disappointed expectations. Pricing is set at $2 per 1M input tokens and $6 per 1M output tokens.
Coverage timeline
Techmeme
xAI : SpaceXAI releases Grok 4.7, which it says is better at verifying its own work and managing longer context, available for $2/1M input and $6/1M output tokens — SpaceXAI's most powerful model for coding and knowledge work. Twice as fast, at half the price of comparable models.

机器之心机器之心
今天一早,SpaceXAI 最强模型 Grok 4.7 来了。 发布页开头只有一句极具「攻击性」的定位:面向编程与知识工作的最强模型,速度达到可比模型的两倍,价格只有一半。 这次升级没有停留在跑分表上。Grok 4.7 换用了更大的基础模型,强化了长时间任务、自我检查和长上下文管理能力,并把文档、演示文稿、法律、医疗和工程等专业工作纳入重点测试。它瞄准的场景很明确:让模型在持续数小时的任务里少走弯路,交出可以直接使用的结果。 马斯克发推表示,「Grok 4.7 在智能水平、运行速度和成本之间取得了很有竞争力的平衡。」 长任务,成为这一代模型的主战场 Grok 4.7 采用了比 Grok 4.6 更大的基础模型。训练阶段延长了强化学习周期,任务组合也进一步加难,其中更多样本需要数小时才能完成。SpaceXAI 称,新模型在检查自身工作、管理长上下文方面都有提升。 这类改进直接对应当前编程智能体最棘手的问题。短代码生成往往只需要局部判断,真正复杂的软件任务却包含阅读仓库、拆解需求、修改多处文件、运行测试和反复纠错。模型能否在漫长执行链中保持目标、发现错误,通常比一次写出漂亮代码更重要。 CursorBench 4.0 专门考察长时间编码任务。官方数据中,Grok 4.7 得分 46.3%,Grok 4.6 为 40.4%,提升 5.9 个百分点。在 DeepSWE v1.1 上,Grok 4.7 的高推理强度成绩为 71.0%;Terminal-Bench 4.0 则从上一代的 20.3% 升至 38.0%。 据此将 Grok 4.7 称为 CursorBench 4.0 上价格与性能均处于前沿的模型。 模型也被训练为原生理解 Grok Bot 的运行框架。按照官方说法,这项调整改善了对话任务和通用知识工作表现。换句话说,Grok 4.7 的升级重点已经从单轮回答延伸到模型与工具、执行环境之间的持续协作. 从写代码,走向完整的知识工作 编程仍是 Grok 4.7 最醒目的标签,但 SpaceXAI 给出的评测范围明显更宽。AA Briefcase 和 GDPval 关注的是专业人士日常完成的多步骤工作,覆盖律师、护士、金融分析师等岗位常见任务,也包括文档和演示文稿制作。 在 AA Briefcase v1.1 上, Grok 4.7 获得 1657 分 ,高于 Grok

Related stories
SpaceXSI launches Grok Voice 3.0, claims top speech-to-speech index spot at $0.08/min
SpaceXSI announced Grok Voice 3.0, a speech-to-speech model it claims tops Artificial Analysis' index with sub-second response times and a price of $0.08 per minute, undercutting rivals. The company describes it as the world's most intelligent voice model, though no independent verification was cited.
1 reports · 1 sources
SpaceXSI unveils unified X Subscription bundling Grok, Cursor, X Premium
SpaceXSI announced a unified membership scheme called the X Subscription, bundling access to Grok, Cursor, and X Premium across its products. The subscription offers four tiers at $8, $30, $100, and $200 per month, with the $30 tier including Grok Bot.
1 reports · 1 sources
Microsoft Preorders Nvidia-Powered Surface RTX Spark Dev Box at $5,999
Microsoft's Nvidia-powered Surface RTX Spark Dev Box is now available for preorder at $5,999, with shipments slated for November. The device features a flat 3D-printed anodized aluminum chassis that doubles as a heatsink, runs on Nvidia's Arm-based RTX Spark platform with 128GB of unified memory, and launches alongside the new Surface Laptop Ultra.
1 reports · 1 sources

Musk: Grok Bot to Use Best Backend Models, Including Rivals' Claude, MidJourney, Suno
Elon Musk announced on Tuesday night that xAI's Grok Bot will use the best backend model for each task, including competitor models such as Claude Opus 5.5, MidJourney, and Suno. This marks a strategic shift for xAI, moving away from relying solely on its own models.
1 reports · 1 sources
News updates every hour. Subscribe in your RSS reader to get new stories automatically.
Subscribe