On 26 October 2026, OpenAI introduced GPT-6.1 Terra, a new model built mainly for work on the computer: filling in forms, reading technical drawings, editing videos and drafting text. What stands out is less a new record than the pricing. According to OpenAI, standard API prices are 50 percent lower than those of GPT-5.6 Terra and GPT-6.1 Sol.
This article puts the launch in perspective: what OpenAI states, what remains a vendor claim and what developers should test now.
Key Takeaways
- OpenAI positions the model as a fast, low-cost generalist for computer tasks with strong visual understanding.
- The API costs $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens.
- All benchmarks cited come from OpenAI itself. Independent comparisons are not yet available.
- For the hardest coding tasks, OpenAI still points to GPT-6.1 Sol and GPT-6 Astra.
- In cybersecurity, OpenAI rates the model “Critical” and applies the same safeguards as for Sol and Astra.
GPT-6.1 Terra at a Glance: What OpenAI Announced
According to OpenAI’s announcement, the model combines detailed visual reasoning with precise computer use and “clearer, more natural writing”. It is a significant step up from GPT-5.6 Terra. OpenAI names tasks that fill the workday as its target, such as drafting proposals, interpreting technical drawings, filling out forms or editing videos.
For developers, the company gives a concrete example: reading incoming invoices and entering the details into accounting software. High-volume jobs like this depend on the price per task, not only on the quality of a single answer.
Pricing and Availability
Starting now, GPT-6.1 Terra is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. Developers can reach it through the OpenAI API as gpt-6.1-terra. Standard prices at a glance:
- Input: $1 per million tokens
- Cached input: $0.05 per million tokens
- Output: $5 per million tokens
OpenAI says that is 50 percent less than GPT-5.6 Terra and GPT-6.1 Sol. In ChatGPT Work and Codex, the same usage allowance covers more tasks than with GPT-5.6 Terra and substantially more than with GPT-6.1 Sol or GPT-6 Astra. Other providers are also pushing prices down, as our article on Kimi K3.1 Pro shows.
Benchmarks: What the Numbers Show and What They Don’t
OpenAI measures computer-use skills with “Agents’ Last Exam”, a set of professional tasks that includes turning architectural drawings into a 3D model or reviewing an animated film shot by shot. The company reports:
- At medium reasoning effort, GPT-6.1 Terra scores 15 percentage points higher than GPT-5.6 Terra at about one-third of the cost.
- That matches GPT-6 Astra at only one-tenth of its price per token.
- At max effort, it is over three times as token efficient as Claude Opus 5.5 at one-fourth of the price per token, which OpenAI translates into roughly 9x lower cost per task.
- Tasks finish more than three times faster than with GPT-6 Astra and GPT-6.1 Sol.
In coding, Terminal-Bench 4.0 rises to 50.3 percent at max effort. GPT-5.6 Terra had a published score of 21.5 percent, according to OpenAI. For the most complex coding tasks, GPT-6.1 Sol and GPT-6 Astra remain the better choice. On Terminal-Bench-Science 0.1, Terra at max effort surpasses GPT-6.1 Sol at lower cost and comes within five points of Opus 5.5 at less than one-fifth of the estimated cost.
The context matters: these are results on benchmarks the vendor selected and ran itself. Only a test on your own workflows shows how the model really performs.
Fewer Errors: What OpenAI Says About Factuality
OpenAI says it has invested heavily in reducing hallucinations. On an internal benchmark of error-inducing prompts, the error rate at low reasoning effort fell from 53.6 percent with GPT-5.6 Terra to 16.7 percent with GPT-6.1 Terra. At maximum effort, it fell by 18 percentage points to 4.3 percent, close to GPT-6 Astra and GPT-6.1 Sol. The model is also said to stay within scope more reliably and to disclose limitations and blockers more honestly. In prompt-injection tests, which use malicious instructions meant to divert the model from the user’s task, OpenAI reports strong resistance.
The test is internal and cannot be verified publicly. It uses deliberately error-inducing prompts and does not describe everyday error rates.
Safety: A “Critical” Rating for Cyber
The safety assessment is notable. According to the system card addendum, OpenAI treats GPT-6.1 Terra under its Preparedness Framework as “Critical” in cybersecurity and “High” in biological and chemical capability. In AI self-improvement it is below the “High” threshold. The same safeguards stack as for GPT-6.1 Sol and GPT-6 Astra applies, with probes updated to be compatible with the model.
The system card reports these measurements:
- ExploitBench: 97.10 percent normalized mean score. OpenAI itself points to possible contamination because the test uses vulnerabilities going back to 2023.
- ExploitBench, internal port (June to August 2026): 24.5 percent arbitrary code execution on recently disclosed vulnerabilities, versus 21.5 percent for GPT-6.1 Sol and 31.5 percent for GPT-6 Astra.
- SEC-Bench Pro: 83.33 percent pass@1 at finding vulnerabilities in large JavaScript engines such as V8 and SpiderMonkey.
OpenAI concludes that reliable exploitation of freshly disclosed vulnerabilities remains challenging for the model. For anyone running web infrastructure, it is still a reminder to apply patches quickly. How fast a flaw becomes a problem was shown recently by the Brotli vulnerability in Node.js.
What Developers Should Do
- Measure it yourself: A pilot with your own data, such as invoices or forms, says more than any vendor table.
- Count cost per task: For agent workflows, token use, speed and error rate matter together, not the raw token price.
- Pick the model by task: For the hardest coding work, OpenAI itself still names Sol and Astra.
- Secure your agents: Computer use needs narrow permissions and oversight. How risky missing oversight can be is shown in our article on AI agents that disable their own monitoring.
The full measurements are in the addendum to the GPT-6 Astra system card.
Conclusion
GPT-6.1 Terra is above all a price and speed offer for computer tasks at scale. The numbers are notable, but they come from OpenAI alone. Independent tests and your own pilots will decide whether the promises hold up in daily work. We will watch what developers report over the coming days.
