OpenAI released GPT-6.1 Terra on 26 October, and the release page is unusually explicit about what kind of model this is. It is not the flagship and it is not pitched as one. It is a workhorse — the model you point at the queue of paperwork that sits between a request and a finished task.
The Pitch Is Cost, Not Capability
Terra’s standard API prices are 50% lower than GPT-5.6 Terra and GPT-6.1 Sol: $1 per million input tokens, $0.05 per million cached input, $5 per million output. In ChatGPT Work and Codex it is included for Plus, Pro, Enterprise and Edu users, with the same allowance covering substantially more tasks than GPT-5.6 Terra.
Read the benchmarks in that light. On Agents’ Last Exam — professional tasks such as turning architectural drawings into a 3D model — Terra at medium reasoning effort scores 15 percentage points above GPT-5.6 Terra at about one-third the cost. It matches GPT-6 Astra at one-tenth the price per token. On MMMU-Pro, visual reasoning, it beats GPT-5.6 Terra at every reasoning setting and nearly matches GPT-6 Astra and GPT-6.1 Sol. On OSWorld 2.1 it surpasses GPT-5.6 Terra’s best score “at a fraction of the cost per task.”
The Claude comparison is harsher still: at max effort, Terra is more than three times as token efficient as Claude Opus 5.5 at a quarter of the price per token, which OpenAI converts into roughly 9x lower cost per task.
Fewer Errors, Not More Knowledge
The factuality numbers may matter more than any of the above. On OpenAI’s internal benchmark of error-inducing prompts, the error rate at low reasoning effort falls from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra. At max effort it drops 18 points to 4.3%, nearly matching GPT-6 Astra and GPT-6.1 Sol. OpenAI frames this as the year’s hallucination work paying off, and the framing is defensible: a model that is cheap and verifiable is worth more in an invoice-processing pipeline than an expensive one that needs a reviewer anyway.
Coding Gets Its Own Jump
Terminal-Bench 4.0: Terra solves 50.3% of tasks at max reasoning, more than double GPT-5.6 Terra’s published 21.5%. That is a large jump for a mid-tier model — but OpenAI’s own caveat is that GPT-6.1 Sol and GPT-6 Astra remain better on the most complex coding tasks, so the flagship tier still has a job. On Terminal-Bench-Science 0.1, Terra beats GPT-6.1 Sol at lower cost and lands within five points of Opus 5.5 at under a fifth of the estimated cost.
Safety and Availability
Terra meets the Critical threshold in cybersecurity under OpenAI’s Preparedness Framework and keeps the protections built for GPT-6 Astra. OpenAI also reports better scope discipline than GPT-5.6 Terra, and strong resistance to prompt-injection attempts that try to divert the model away from the user’s task. Details are in the system card.
The API model ID is gpt-6.1-terra, available immediately.
The structural story is the one worth carrying forward: OpenAI is not just shipping a cheaper model, it is shipping a model that is cheap because the accuracy work landed. That combination is what turns a benchmark win into a margin change in someone else’s business.
Related: Introducing GPT-6.1 Terra Related: GPT-6.1 Terra system card