OpenAI released GPT-6.1 Terra on Monday, a mid-tier model priced at 1 dollar per million input tokens and 5 dollars per million output tokens, with cached input at 5 cents per million. OpenAI says that is 50 percent lower than both GPT-5.6 Terra, the model it replaces, and its larger GPT-6.1 Sol. The model is live in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu subscribers, and in the API as gpt-6.1-terra, with a context window of about 1.05 million tokens and up to 128,000 output tokens. It is not offered in ordinary ChatGPT conversations.
Terra is built for the workday, not the leaderboard: drafting proposals, reading technical drawings, filling out forms and editing video, pairing detailed visual reasoning with what OpenAI calls precise computer use and clearer, more natural writing. On the company's own figures, it scores 15 percentage points higher than GPT-5.6 Terra on Agents' Last Exam at medium reasoning effort at about a third of the cost, solves 50.3 percent of Terminal-Bench 4.0 coding tasks at maximum effort against 21.5 percent for its predecessor, and cuts a low-effort factual error rate from 53.6 percent to 16.7 percent. OpenAI also says it finishes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol.
The independent check is more measured, and still favorable. Artificial Analysis found Terra gains 6 points on its Intelligence Index over GPT-5.6 Terra at maximum effort while costing less than a third per task, 0.40 dollars against 1.40 dollars, with gains of 7 to 13 points at lower effort levels and low and medium settings sitting on the cost-efficiency frontier. It also reported the model guessing 38 percent of the time when it does not know an answer in its AA-Omniscience test, down from 88 percent for the predecessor. The rest of the launch figures are OpenAI's own and have not been independently tested, so treat speed and computer-use claims as claims until your own workload confirms them.
For builders, cost per task is now the margin. Cloudera's State of Enterprise AI 2026 report, released the same day from a survey of 1,500 IT leaders, found 89 percent had already limited or redesigned AI use cases because of rising token consumption. A capable mid-tier at half price, with cached context at a quarter of the old rate, changes the default: route routine agent loops, document work and desktop automation to Terra-class models first, cache aggressively, and reserve a flagship for the hardest steps only. Watch the fine print before migrating volume, though: requests over 272,000 input tokens bill at higher rates across the whole request.
The timing is the tell. Terra landed the same day Anthropic filed publicly for its IPO, and OpenAI pointedly compared its per-task cost with Claude Opus 5.5, claiming about one-ninth the cost at maximum effort on its own calculations. Whether or not that comparison survives contact with real workloads, the contest has clearly moved: not who has the smartest model at any price, but who delivers a finished, checked task for the fewest cents. Builders who instrument cost per completed task, not cost per token alone, will know first whether Terra actually wins it.
