GPT-6.1 Terra's price cut: what it means for your business

OpenAI shipped GPT-6.1 Terra on October 26, 2026, as a faster, cheaper tier sitting between its flagship reasoning models and its older agent models. The headline change is price: $1 per million input tokens and $5 per million output tokens on the API, which OpenAI says is half of what GPT-6.1 Sol and the model it replaces, GPT-5.6 Terra, cost. It also runs noticeably faster on computer-use tasks, the kind where a model has to click through a real screen or app rather than just answer a question in text.

Terra isn't the model OpenAI recommends for the hardest coding or research work, that's still Sol or GPT-6 Astra. It's built for the high-volume, repeatable agent work a business actually runs thousands of times a day once the price is low enough: filling out forms, moving data between systems, clicking through a familiar web app on a schedule.

Not directly, and that's the detail most coverage of the launch skipped. GPT-6.1 Terra rolled out in ChatGPT Work, OpenAI's workspace built around letting an agent act inside real apps, and in Codex, its coding assistant. It did not roll out to the regular ChatGPT Chat interface that most Plus, Business, and Enterprise seats use day to day. If your team's whole relationship with ChatGPT is typing questions into chat, this release doesn't touch your bill or your experience yet.

The price cut lands on OpenAI's API, the metered, pay-as-you-go usage that powers custom tools, chatbots, and automations, not a flat seat price. If your business, or a vendor or developer you pay, has built something on OpenAI's API that does agent-style work (moving data, filling in forms, clicking through a browser on your behalf), and it's already calling a Terra-class model, that work got roughly half as expensive to run with no code change required.

If it was built on a pricier model like Sol or Astra because the task genuinely needs deeper reasoning, that calculation hasn't changed. Terra is positioned as the mid-tier option, not a replacement for the top one. The cut mainly matters for the agent and computer-use work that was already a fit for a lighter model.

Two concrete moves. First, if an agency, developer, or in-house tool already runs agent tasks on OpenAI's API, ask which model it's pointed at, and whether swapping to gpt-6.1-terra holds up on quality across a sample of real requests before changing the production default. That swap is usually a one-line model ID change, not a rebuild. Second, if a repetitive, rules-based task (data entry, form filling, routine browser steps) has been sitting on a 'too expensive to automate' list because of per-token cost, this is the price point worth testing it against again.