GPT-6.1 Terra: what OpenAI showed and what it changes for your company
OpenAI launched GPT-6.1 Terra with API prices 50% lower and fewer errors. We redrew the announcement's charts and explain what matters.
· Josué Juca
- GPT-6.1 Terra launched on October 26, 2026 and costs US$1 per million input tokens and US$5 per million output tokens.
- The standard API price is 50% lower than GPT-5.6 Terra and GPT-6.1 Sol.
- According to OpenAI, the error rate on error-inducing prompts dropped from 53.6% to 16.7% (low effort) and the programming score more than doubled.
- The numbers are OpenAI's own. Test with your own documents before deciding.
On October 26, 2026, OpenAI introduced GPT-6.1 Terra, described as a fast and efficient model that combines detailed visual reasoning and precise computer use with clearer, more natural writing. The promise is a new balance of speed, capability and cost, and the company cites as an example of use automating high-volume tasks, such as reading incoming invoices and entering the data into accounting software.
In this article we redrew the main charts from the presentation, using the numbers OpenAI published, and translated what they mean for anyone automating processes.
- per million input tokens
- US$1
- per million output tokens
- US$5
- lower than the standard price of GPT-5.6 Terra and GPT-6.1 Sol
- 50%
Fewer factual errors
OpenAI says it invested heavily this year in reducing hallucinations and factual errors. On an internal benchmark of error-inducing prompts, the error rate dropped sharply compared with GPT-5.6 Terra:
Error rate on error-inducing prompts (lower is better)
- Low reasoning effort: GPT-5.6 Terra 53.6%; GPT-6.1 Terra 16.7%
- Max reasoning effort: GPT-5.6 Terra 22.3%; GPT-6.1 Terra 4.3%
Source: OpenAI, internal benchmark. At max effort, OpenAI reports a drop of 18 percentage points to 4.3%; the 22.3% value for GPT-5.6 Terra was calculated from that.
In practical terms, at max effort Terra gets close to the rates of GPT-6 Astra and GPT-6.1 Sol. For anyone using AI to read documents or answer customers, making fewer mistakes is what weighs the most.
Programming and terminal tasks
On Terminal-Bench 4.0, which tests whether agents can solve command-line problems, such as fixing a web page whose layout shifts while loading, the jump is large:
Terminal-Bench 4.0, max effort (% of tasks solved)
- GPT-5.6 Terra: 21.5%
- GPT-6.1 Terra: 50.3%
Source: OpenAI. The GPT-5.6 Terra score is the one previously published.
OpenAI itself adds a caveat: for the most complex programming tasks, GPT-6.1 Sol and GPT-6 Astra remain the best options.
Computer use at a lower cost
On Agents' Last Exam, a benchmark of professional tasks that includes turning architectural drawings into a 3D model, OpenAI highlights these results:
- At medium effort, Terra scores 15 percentage points higher than GPT-5.6 Terra, for about a third of the cost.
- It even matches GPT-6 Astra for only a tenth of the price per token.
- At max effort, Terra is more than three times more token-efficient than Claude Opus 5.5 and costs a quarter of the price per token, which gives a cost per task about 9 times lower, according to OpenAI.
- Tasks finish more than three times faster than on GPT-6 Astra and GPT-6.1 Sol.
All these comparisons come from OpenAI's own tests. Benchmark results help compare models, but they do not replace a test with your own documents.
Pricing and availability
Standard API price per million tokens (US$)
- Cached input: US$ 0.05
- Input: US$ 1
- Output: US$ 5
Source: OpenAI. The model name in the API is gpt-6.1-terra.
GPT-6.1 Terra is available to all users on Plus, Pro, Enterprise and Edu plans, in ChatGPT Work and Codex. OpenAI states that, on those plans, the same usage limit allows more tasks with Terra than with GPT-5.6 Terra, GPT-6.1 Sol or GPT-6 Astra.
Safety and alignment
According to OpenAI, the model reaches the Critical cybersecurity threshold of its Preparedness Framework and keeps the safeguards developed for GPT-6 Astra. In the company's evaluations, it also stays more within scope, communicates its limitations more honestly and resists malicious instructions hidden in a text better (prompt injection). Details are in the system card published by OpenAI.
What it changes for your company
When the cost per document drops and the error rate does too, tasks that were not worth it before start to make sense. A few examples close to what we do with our clients:
- Reading invoices, contracts and PDFs and moving the data into spreadsheets and systems.
- Classifying emails and tickets and triggering the right workflow.
- Drafting reports from your data for a person to review.
- Filling in forms and portals, combining AI with RPA.
How to evaluate before adopting
- Test with a real sample of your documents, not demo examples.
- Compare cost per document and error rate with what you use today.
- Keep human review on critical cases, because any model can be wrong.
- Check where your data goes and how that fits with LGPD.
- Avoid depending on a single model: names, prices and versions change often.
If you want help measuring that gain in your process, we can set up a small pilot and integrate AI with your systems.
Sources