OpenAI has launched a new model, GPT-6.1 Terra, positioned as the "fast and efficient mid-tier". It bundles fine-grained visual reasoning, precise computer-operation capability, and more natural writing together, targeting high-frequency office tasks like drafting proposals, reading technical drawings, filling out forms, and editing video. Compared with the previous-generation GPT-5.6 Terra, the company calls it a clear leap forward.
Mid-tier positioning, yet it can "get hands-on"
Terra is now available to ChatGPT Plus, Pro, Enterprise, and Edu users, embedded in ChatGPT Work and Codex, and developers can also call it via the API as gpt-6.1-terra. Its most eye-catching label is "autonomous computer operation" — on the desktop-operation benchmark OSWorld 2.1, at medium reasoning effort it already surpassed GPT-5.6 Terra's best result.
Cost-performance is another selling point. On the Agents' Last Exam benchmark, at medium effort Terra scored 15% higher than GPT-5.6 Terra while costing only about one-third per task, and its results were on par with the flagship GPT-6 Astra, at just one-tenth of the price per token. At full effort, its token efficiency is more than three times that of Claude Opus 5.5 at roughly one-quarter the unit price, working out to a per-task cost about 9 times lower, and it completes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol.
Sweeps the old model on every benchmark, safety raised to Critical
On nearly every benchmark it crushes the old model: on visual reasoning, MMMU-Pro, it exceeds GPT-5.6 Terra at every effort level, approaching the two flagships at lower cost; on factuality testing, the error rate at low effort dropped from 53.6% to 16.7%, and was pressed further down to 4.3% at high effort; on the coding benchmark Terminal-Bench 4.0 it solved 50.3% of tasks at high effort, more than double the old model's public result.
As capabilities rose, the guardrails were raised too. Terra reached the Critical level in OpenAI's own cybersecurity evaluations, inheriting GPT-6 Astra's safeguards, better at holding task boundaries and proactively disclosing limitations, with improved resistance to prompt injection as well.