GPT-6 Astra 与 Opus 5.5 每任务成本对比
帖子按 Artificial Analysis 数据算每完成任务成本：GPT-6 Astra 约 1.35 美元、Opus 5.5 约 2.38 美元，并提到 Astra 是 OpenAI 助理 Dots 的底层模型。
Sam Altman's new model is 2.5x more expensive than Claude. And it still costs you less. GPT-6 ASTRA: $10 PER 1M INPUT. OPUS 5.5: $4. PER TASK ASTRA WINS $1.35 VS $2.38 The trick is how much each model talks. Output tokens per task (Artificial Analysis run): GPT-6 Astra: ~27K Opus 5: ~73K Fable 5.1: ~78K Opus 5.5: ~119K Opus 5.5 writes 4x more to finish the same job. You pay for every one of those tokens. Input price per 1M, from the chart: Jev: $0.042 Haiku 5.5: $0.1 Sonnet 5.5: $2 Opus 5.5: $4 GPT-6 Astra: $10 So the per-token price tag lies. Price per finished task is the number that hits your bill. Where Astra stops being cheap: → past 272K input tokens the whole request bills at 2x input → output is $50 per 1M, so long answers hurt → cached input drops to $1, reuse your context Astra is also the brain inside Dots, OpenAI's agent with its own cloud computer, a browser and 4,000+ apps. First Dot comes with Pro and Business Premium. I check cost per task now. Per-token pricing told me the wrong model. Full price map is in the image. Save it before you pick a model by its sticker price.
这条是引用着另一条写的：x.com/donatelloxpro/status/2109320196649619552