The math on AI automation just changed
OpenAI's new Sol model matches its best AI at one-fifth the price, making serious automation affordable for teams that couldn't justify the cost before.
- $2 / $10
- Sol price per million tokens (in / out)
- ~1/5th
- Sol's cost compared to Astra
- 4,000+
- apps Dots agents connect to
- 35%
- agent task success rate without human help
For most of 2026, running a powerful AI model on real work — not a demo, but a system handling tasks every hour — meant high costs. OpenAI's flagship model, GPT-6 Astra, charges $10 per million words you send it and $50 per million words it writes back. For a team automating serious work, that adds up fast. On October 6, OpenAI launched Sol at $2 in and $10 out — one-fifth of Astra's price — and that changes the calculation entirely.
Sol is not a cut-down, slower model. It matches Astra's performance on coding, research, and professional tasks — just at a fraction of the cost. For a small team running an AI tool around the clock, that is the difference between a bill that hurts and one that fits your budget. The price of serious AI automation just fell to where most businesses can actually afford it.
OpenAI also launched "Dots" at the same event: personal AI agents that run all day in the background, watch your apps, and take actions across more than 4,000 tools. Right now, Dots runs on the expensive Astra model. But Sol points the direction — as frontier models get cheaper, always-on agents move from an enterprise luxury to something any team can run without a huge budget.
Every new top AI model starts expensive. Then a cheaper, nearly-as-good version follows. Sol fits that pattern exactly — it already matches Astra on coding benchmark tests. If you have been waiting to build something real with AI because the costs felt too high, now is the moment to run a real test on Sol and see what the numbers look like for your actual work.