OpenAI released GPT-6.1 Sol on September 29 at $2 per million input tokens and $10 per million output tokens. That is exactly one-fifth of what GPT-6 Astra costs. On a benchmark built from real software bug fixes, Sol scored 75.2 percent against Astra's 74.8.
Hand it the same coding work you give Astra today. If your team writes scripts, builds forms, fixes bugs, or generates API calls with AI, switch the model to Sol in your next session. The price drop is real: a task that cost you $50 with Astra now runs for $10. Tasks that felt too expensive to automate last month might make sense this week.
You can also try it for customer-support replies, proposal drafts, or anything that used to blow through your monthly budget before the job was done. Run it against your normal work, not a demo. If the output holds up, keep it. If it does not, you spent $10 finding out.
A person still reads every line of code before it goes live. The cheaper price does not mean you trust it more. Set a rule: no AI code ships until someone on your team reads it, understands what it does, and decides it is safe. Keep a log of what the model built and who approved it. If the task involves customer data, payment workflows, or access rules, have two people check it. That is the line. Faster and cheaper are great. Unreviewed is not.