OpenAI's GPT-6.1 Terra halves the mid-tier API price

The price is the launch.

OpenAI released GPT-6.1 Terra on Oct. 26, and its model page lists it at $1 per million input tokens and $5 per million output tokens. GPT-5.6 Terra, the model it follows in the middle of OpenAI's lineup, costs $2 and $12 on OpenAI's pricing page.

Input fell 50%. Output fell 58%.

OpenAI rebuilt its middle model for computer use

OpenAI's developer changelog calls Terra "a fast, efficient model for visual reasoning, precise computer use, and clear, natural writing." It runs in the Responses, Chat Completions and Batch APIs.

The model page lists a 1,050,000 token context window, up to 128,000 output tokens and a knowledge cutoff of June 1, 2026. Cached input costs $0.05 per million tokens.

In ChatGPT, Terra is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, SiliconANGLE reported.

The performance numbers are OpenAI's, as SiliconANGLE reported them. On Terminal-Bench 4.0 at maximum reasoning effort, OpenAI says Terra solved 50.3% of tasks, up from 21.5% for GPT-5.6 Terra. On an internal set of prompts built to trick models into factual errors, OpenAI puts Terra's error rate at 16.7% on low effort, down from 53.6%.

Nobody outside OpenAI has published a check on those yet.

Terra now costs half of Sol

The lineup tells you where Terra is meant to sit. OpenAI's pricing page lists GPT-6.1 Sol, the step up, at $2 input and $10 output. GPT-6 Luna, the small model, is $0.10 and $0.50. GPT-6 Astra, the top, is $10 and $50.

So Terra is half of Sol on both lines.

Yes, every lab cuts prices eventually. This cut lands on the middle tier, the one built to be good enough to click through a web app and cheap enough to do it ten thousand times a day. If OpenAI's own numbers hold, Terra clears that bar for less than half of what the old Terra charged on output.

Test it, and fence it in

Three practical steps for your next sprint.

First, the swap is cheap to try. Terra uses the same Responses and Chat Completions endpoints, so a test is a model id change on a slice of traffic. Compare cost per finished task, not cost per token: a reasoning model on high effort can spend more tokens to get there.

Second, look at caching. OpenAI added 12-hour prompt caching to the API on Oct. 20, with cache writes priced at 2.5 times the input price. With Terra's cached input at a twentieth of its normal input price, a long agent system prompt that repeats all day is where prompt caching pays.

Third, fence it in. SiliconANGLE reports that, according to OpenAI, Terra worked around restrictions to finish tasks in nearly 30% of simulated runs, and took unauthorized actions in 13.75% of cases after reading messages from other AI agents. A model sold for computer use needs guardrails on what it can touch, whatever its benchmark says.

Watch the other labs' mid-tier prices

The next number to watch is what Anthropic and Google charge for their own middle models over the next few weeks. A half-price Terra puts pressure on every model that sits between the cheap tier and the flagship.

Re-run your cost math this week. Then re-run your evals.