GPT-6 Astra for stock trading

GPT-6 Astra is OpenAI's top model, and since DevDay on September 29, 2026 it is also the model inside every dot, OpenAI's always-on agent. The API model ID is gpt-6-astra. It works with Chat Completions, Responses, and Batch. Realtime, Assistants, and fine-tuning are not supported.

The price is the first thing a trading agent runs into. Input is $10 per million tokens, cached input is $1, and output is $50. A request over 272K input tokens costs double on input and 1.5x on output for the whole request, so a loop that keeps stuffing fill history into the prompt pays a penalty once it crosses that line.

Here is a rough cycle. The system prompt and skill file are 20K tokens and cache after the first call. Quotes and portfolio add 3K fresh tokens. The model writes 2K tokens of reasoning and orders. That is about $0.15 a cycle. Every 30 minutes through a US session is 13 cycles, or about $2 a day. Every 5 minutes around the clock for crypto is closer to $43 a day. Reasoning effort goes from low to max, and max will push output well past 2K.

Most operators will not put Astra in the trade loop. GPT-6.1 Sol, announced at the same DevDay, is $2 in and $10 out and scores within a point of Astra on Artificial Analysis's index. The pattern that pays is Astra for the weekly strategy review, where the 1M window can hold months of your own fills and reasoning, and a cheaper model for the minute-to-minute calls.