OpenAI says its new GPT-⁠6.1 Terra runs three times faster than Sol at half the price

OpenAI released GPT-6.1 Terra for computer use

OpenAI announced GPT-6.1 Terra on Oct. 26. The company describes it as a fast, efficient model that combines detailed visual reasoning and precise computer use with clearer, more natural writing. Terra understands what is on screen and takes action across apps. OpenAI’s examples include drafting proposals, interpreting technical drawings, filling out forms and editing videos.

In OpenAI’s lineup, Terra ranks under GPT-6 Astra, its most capable model, and GPT-6.1 Sol, with Luna a tier lower. It replaces GPT-5.6 Terra. In the launch example OpenAI posted on X, the model builds a detailed 3D model from a photo of a tram.

The model is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and in the API as gpt-6.1-terra. According to Decrypt, free users do not get Terra and stay on GPT-5.6 Luna.

OpenAI@OpenAI

GPT-6.1 Terra: delivering fast and capable computer use at a lower price.

GPT-6.1 Terra understands what’s on screen, takes action across apps, writes naturally, and is a more capable model for your whole workday with the efficiency to take on more of it.

Pricing: half off on input, more than half off on output

Standard API pricing is $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. OpenAI says that is 50% lower than GPT-5.6 Terra and GPT-6.1 Sol. The Decoder put GPT-5.6 Terra at $2 for input and $12 for output. That means input pricing was halved, output fell about 58% and cached input dropped from $0.20 to $0.05.

By independent evaluator Artificial Analysis’s count, Terra spends $0.40 per task on average at its highest reasoning setting, versus $1.40 for GPT-5.6 Terra. OpenAI says the same usage allowance in ChatGPT Work and Codex also covers substantially more tasks than with Sol or Astra.

Speed and accuracy in OpenAI’s own tests

OpenAI measured computer use on Agents’ Last Exam, a benchmark of professional tasks such as turning architectural drawings into a 3D model. At medium effort, the company says Terra scored 15 points higher than GPT-5.6 Terra at about a third of the cost. It matched GPT-6 Astra at a tenth of the price per token and finished tasks more than three times faster than Astra and Sol. At max effort, OpenAI says its cost per task is about one ninth of Claude Opus 5.5’s.

On OpenAI’s internal benchmark of error-inducing prompts, the error rate fell from 53.6% to 16.7% at low effort and to 4.3% at max effort. On Terminal-Bench 4.0, a command-line coding test, Terra solved 50.3% of tasks at max effort, versus a published 21.5% for GPT-5.6 Terra. OpenAI still recommends Sol and Astra for the most complex coding. Decrypt noted that all of these figures are OpenAI’s own and that the error test is internal.

48 on the Artificial Analysis index, 4 points behind Sol

On Artificial Analysis’s Intelligence Index v4.3.2, built from 10 evaluations, Terra scored 48 at max effort. GPT-5.6 Terra scored 42, GPT-6.1 Sol 52 and GPT-6 Astra 53, with Claude Opus 5.5 on top at 58. The firm said Terra gained 7 to 13 points over its predecessor at lower effort settings.

On the firm’s coding agent index, Terra scored 48 to Sol’s 54. In Artificial Analysis’s own Terminal-Bench 4.0 run, Terra reached 55.6%, just behind Sol at 56.1%, while GPT-5.6 Terra had 35.4%. Kimi K3.1 Pro led that test at 67.2%.

Made-up answers fell from 88% to 38%

Artificial Analysis’s AA-Omniscience test measures how often a model answers incorrectly when it should have declined or admitted it didn’t know. Terra’s rate fell to 38%, from 88% for GPT-5.6 Terra, while its accuracy on the same test rose 2 points.

That is lower than Claude Opus 5.5 (59%), GPT-6.1 Sol (54%) and GPT-6 Astra (51%). It is not the best result, though: Gemini 4 Argon leads at 15%, and eight models rank ahead of Terra.

System card: “Critical” in cybersecurity

According to OpenAI’s system card, the company is treating Terra as “Critical” in cybersecurity, the top level of its Preparedness Framework. The model ships with the safeguards built for GPT-6 Astra. It is rated “High” in biology and chemistry and stays under the High threshold for AI self-improvement.

One finding stands out. At max effort, when the model hit a barrier, it completed the task through a route that violated the spirit of the restriction in 29.6% of runs. That is higher than Astra, Luna and GPT-6.1 Sol. The Decoder reported that its defense against indirect prompt injection rose from 96.2% to 98.8%.