OpenAI's New AI Costs Half as Much — But Claude Still Codes Better

On September 22, OpenAI and Anthropic each released a new AI model built for the same job: writing software and carrying out multi-step tasks on its own (called "agents" — AI that can take actions, not just chat). OpenAI's GPT-6 Sol and Anthropic's Claude Opus 5.5 are direct rivals. The headline difference is price: Sol costs half what Opus 5.5 does for everyday use.

AI bills are measured in "tokens" — chunks of text roughly the size of a short word. Sol charges $2 per million tokens you send it and $10 per million it writes back. Opus 5.5 charges $4 and $20. So a company burning through AI to write code could cut that bill in half by switching. But there's a catch. Once a single request passes about 272,000 tokens — roughly a thick novel — Sol jumps to the same $4 rate, and some reused text actually costs twice as much. Anthropic keeps one flat rate even at a million tokens.

The two also differ in how much "thinking" you can control. Sol offers six settings, including one that turns extended thinking off entirely — handy when you want fast, cheap answers. Opus 5.5 offers four and can't switch thinking off. In exchange, Claude wins several current coding tests, especially on long, complicated software jobs. Both can read about a million tokens at once, which is like handing the AI an entire company codebase in one go.

For you, the practical effect is cheaper AI tools. Price wars between OpenAI and Anthropic usually show up as lower subscription costs and better free tiers within months. It also means more companies can afford to automate work — great for speed, worth watching if your job involves routine coding or document review. The simple rule for teams: pick Sol for everyday jobs, Claude when the task is long and messy.