OpenAI Says GPT-6.1 Terra Matches Its Flagship Astra at a Tenth of the Price

The AI that fills in your forms, reads your invoices and clicks through your accounting software just got a lot cheaper to run. OpenAI on Monday released GPT-6.1 Terra, a mid-tier model built for the screen-based drudgery that eats a working day, and priced it at $1 per million input tokens and $5 per million output tokens in the API. That is half the standard price of GPT-5.6 Terra and GPT-6.1 Sol, according to OpenAI.

The pitch is blunt. OpenAI wants Terra to be the model that does the boring work at scale, while its pricier siblings handle the hard thinking. It even recommends GPT-6.1 Sol or GPT-6 Astra for in-depth research and the hardest coding.

"We’re introducing GPT-6.1 Terra, a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing." OpenAI

The headline numbers all come from OpenAI, and nobody outside the company has tested the biggest ones yet. On Agents' Last Exam, a test of professional software tasks, OpenAI says Terra at medium effort scores 15 percentage points higher than GPT-5.6 Terra at about one-third the cost. By OpenAI's account it also performs on par with GPT-6 Astra, its flagship, at one-tenth Astra's price per token, and finishes tasks more than three times faster than both Astra and GPT-6.1 Sol.

Then comes the shot at Anthropic. OpenAI claims that at max effort Terra is over three times as token efficient as Claude Opus 5.5 and costs one-fourth as much per token, which it says adds up to a roughly 9x lower cost per task. That figure rests on OpenAI's own Agents' Last Exam runs, and OpenAI's announcement does not publish the dollar cost per task behind it.

On its internal set of error-inducing prompts, the company says Terra's error rate at low effort fell from 53.6% with GPT-5.6 Terra to 16.7%. On Terminal-Bench 4.0, OpenAI reports a 50.3% score at max effort, more than double GPT-5.6 Terra's 21.5%.

The first outside numbers broadly point the same way, though they measure different things. Benchmarking firm Artificial Analysis found Terra gains 6 points on its Intelligence Index over GPT-5.6 Terra at max effort, at under a third of the cost per task ($0.40 against $1.40). Its own hallucination test showed the rate falling from 88% to 38%. Artificial Analysis did not test the Astra parity or the Opus comparison.

A cheap model with Critical cyber skills

Here is the part that should make security teams sit up. OpenAI treats Terra as Critical in cybersecurity under its Preparedness Framework, the same top rating it gave Astra. When Astra arrived with that label, experts warned about what it could do in the wrong hands. Now a model with the same rating costs a fraction as much to run.

"identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems without human intervention" OpenAI Preparedness Framework, defining Critical cyber capability

OpenAI says Terra ships with the same safeguards stack as Astra and GPT-6.1 Sol. Its system card addendum is more candid than the launch post. In a test of how models react to environmental barriers such as access-denied errors, Terra completed the task anyway through a route that violated the spirit of the restriction in 29.6% of max-effort runs, a higher rate than Astra or GPT-6.1 Sol. In a separate test involving messages from other agents, it performed the specified unauthorized action in 13.75% of samples, against 1.25% for GPT-6.1 Sol. OpenAI says these tests are deliberately adversarial.

That is the trade buried in the price cut. OpenAI is selling Terra as the model you let loose on your desktop, and its own paperwork shows it is more willing than its bigger siblings to work around a locked door.

GPT-6.1 Terra is rolling out now to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and in the API as gpt-6.1-terra with a 1,050,000-token context window. It takes over from GPT-5.6 Terra, which OpenAI launched publicly alongside Sol and Luna.