CodeRabbit 发布的评估报告显示,OpenAI 的 GPT-6 Astra 在代码审查能力上有显著提升1。在简单代码审查场景中,Astra 相比 GPT-5.6 Sol 多捕获约 4% 的可行性 Bug,相比 Opus 5 则多捕获 22%1。在更复杂的跨文件审查中,Astra 的优势进一步扩大,对比 Sol 领先 20%,对比 Opus 5 领先 33%1。
然而,这种性能收益伴随着显著的成本上升1。Astra 的 API 定价为输入令牌每百万 10 美元、输出令牌每百万 50 美元1。在假设固定用量(10 万输入令牌加 1 万输出令牌)的情况下,Astra 的成本是 Sol 的 2.5 倍1。在数据隐私方面,OpenAI 表示符合条件的 API 客户可享受 Astra 的零数据保留政策1,而 Anthropic 的 Fable 则默认要求 30 天数据保留,符合条件的客户才可使用零数据保留选项1。
CodeRabbit released a performance evaluation of OpenAI's GPT-6 Astra model for code review tasks 1. In standard code review scenarios, Astra captured approximately 4% more actionable bugs compared to GPT-5.6 Sol and 22% more than Opus 5 1. The performance gap widened in more complex cross-file reviews, where Astra demonstrated a 20% advantage over Sol and 33% over Opus 5 1.
Pricing represents a significant consideration for users evaluating the model. Astra's API costs $10 per million input tokens and $50 per million output tokens 1. Under a baseline usage scenario of 100,000 input tokens and 10,000 output tokens, Astra's costs are 2.5 times higher than Sol's 1. Regarding data privacy, OpenAI supports zero data retention for qualifying API customers using GPT-6 Astra 1, while Anthropic's Fable model requires 30-day data retention by default, though qualifying customers can opt into zero data retention 1.
评论
还没有评论,欢迎留下第一条。