OpenAI's GPT-6.1 Sol is smarter for the same price, which almost never happens with software
By Brett Gavaghan, Founder of Adaptd ·
OpenAI released GPT-6.1 Sol on 30 September, an upgrade to GPT-6 Sol that scores higher on coding and computer-use tests at the same API price, with repeat prompts now half the cost.
The short version
- OpenAI released GPT-6.1 Sol on 30 September 2026 for ChatGPT and the API, as an upgrade to GPT-6 Sol.
- It costs $2 per million input tokens and $10 per million output in the API, and cached input drops to $0.10.
- If you pay for API usage, test it on your real tasks before switching anything important.
Software updates usually come in two flavours: more expensive, or the button you use every day has moved. GPT-6.1 Sol is the rare third kind.
What is GPT-6.1 Sol?
GPT-6.1 Sol is an updated version of OpenAI's GPT-6 Sol model, released on 30 September 2026 for complex coding and professional work. OpenAI describes it as near-Astra performance at a much lower cost, Astra being its top model.
OpenAI's numbers: 75.2% on the DeepSWE v1.1 coding benchmark at high reasoning effort, up from 68.8% for GPT-6 Sol, and 71.4% on the OSWorld 2.0 computer-use test against Astra's 73.5%. Benchmarks are a bit like a car's fuel figures. Handy for comparing, rarely what you get on the school run.
What does it cost, and who can use it?
GPT-6.1 Sol costs the same as GPT-6 Sol in the API: $2 per million input tokens and $10 per million output tokens. Cached input, the price for text the model has already seen in a repeated prompt, drops from $0.20 to $0.10 per million.
It's rolling out globally to the API and to ChatGPT Work. OpenAI's release notes list it for Business and Enterprise workspaces, and for Pro users first on personal plans. Some developers reported delays getting it in certain desktop apps.
Should your business switch to it?
Businesses that pay for OpenAI API usage should test GPT-6.1 Sol on their own tasks, because a better model at the same price is close to a free upgrade. The cached-input cut helps most where the same long instructions go out again and again, like a support bot.
My advice: don't flip the switch on everything at once. Run a week of your usual prompts through both models and compare the answers. Then check the bill. If you only use ChatGPT through a subscription, there's nothing to do. Occasionally software gets better without asking for your credit card. Enjoy it while it lasts.
Wondering what this means for your business?
That's exactly what the discovery call is for.