SpaceXAI于2026年9月21日发布了推理大模型Grok 4.7（xhigh版本）1。该模型在人工智能指数上获得46分的评分，相比同价位竞品的中位数24分表现更优1。从定价角度看，Grok 4.7具有竞争力，输入令牌定价为每百万2美元，输出令牌为每百万6美元1。
然而，该模型也存在明显不足之处。其输出速度为39.3令牌/秒，低于同价位中位数的72.5令牌/秒1。在评估过程中，模型生成了240万输出令牌，远超同价位中位数的95万1，这反映出其生成文本冗长的特点。模型支持500k令牌的上下文窗口1，首令牌延迟为0.88秒，相比同价位中位数的3.70秒则具有优势1。
SpaceXAI released Grok 4.7 (xhigh), a reasoning-focused large language model, on September 21, 2026 1. The model achieved a score of 46 points on the Artificial Analysis Intelligence Index, substantially outperforming the median score of 24 points among competitors in its price tier 1.
Priced competitively at $2.00 per million input tokens and $6.00 per million output tokens, Grok 4.7 offers an attractive cost structure 1. The model supports a 500,000-token context window and demonstrated a first-token latency of 0.88 seconds, faster than the category median of 3.70 seconds 1. However, the model's output speed of 39.3 tokens per second lags considerably behind the same-tier median of 72.5 tokens per second, and assessment testing revealed a tendency toward verbose text generation 1. During evaluation of the Intelligence Index benchmark, the model generated 240 million output tokens compared to a category median of 95 million 1, underscoring this verbosity issue. The total evaluation cost for the benchmark assessment reached $4,967.35 1.
评论
还没有评论，欢迎留下第一条。