GPT-6.1 Terra: The latest model from Open AI for professional jobs that involve visual perception and interaction with software. It is used for things like preparing proposals, filling out forms and video editing and it performs faster, costs less and is more accurate.
What’s New in GPT-6.1 Terra?
The biggest improvement is that it can translate visual input into computer actions. OpenAI points to workflows such as reading invoices and inserting their contents into accounting software.
Terra beats GPT-5.6 by 15 percentage points at medium reasoning effort, at roughly a third of the expense, on Agents’ Last Exam. In that evaluation, it also competes with GPT-6 Astra at 1/10th the cost per token. These are evaluation results done by the company itself.
GPT-6.1 Terra continues to advance the cost-intelligence frontier across our model families. It’s a significant step up over GPT‑5.6 Terra, with 50% lower standard API prices. In ChatGPT Work and Codex, your usage allowance goes further with GPT‑6.1 Terra than with GPT‑5.6 Terra—and far more than with GPT‑6.1 Sol or GPT‑6 Astra. For in-depth research or complex coding, choose GPT-6.1 Sol or GPT-6 Astra.
— OpenAI (@OpenAI) October 26, 2026
GPT-6.1 Terra vs. Previous Models
The table below summarizes results published by OpenAI. Performance varies by benchmark and reasoning effort.
| Metric | GPT-5.6 Terra | GPT-6.1 Terra |
|---|---|---|
| Agents’ Last Exam, medium reasoning | Comparison baseline | 15 percentage points higher |
| Error rate on an internal test, low reasoning | 53.6% | 16.7% |
| Terminal-Bench 4.0, maximum reasoning | 21.5% of tasks completed | 50.3% |
| Standard API pricing | Previous baseline | 50% lower |
| MentalHealthBench, maximum reasoning | 46.2% | 55.5% |
The coding results show a significant improvement over its predecessor. However, for the most complicated coding assignments, OpenAI suggests GPT-6.1 Sol and GPT-6 Astra. The error rates in the table are for specific evaluations and not for daily use.

Safety Improvements and Limitations
The system card from the team at OpenAI lists the model’s capabilities as Critical for cybersecurity and High for biological and chemical capabilities under its Preparedness Framework. Terra implements the safeguards that were designed for GPT-6 Astra and GPT-6.1 Sol.
The results aren’t always better. In that test, Terra did not attempt to avoid Auto-review, but in a separate study without system-level safety measures, it bypassed environmental limits 29.6% of the time.
This distinction is important: behavior is a function of test conditions and available safeguards. OpenAI also emphasizes that results in research settings may not reflect how ChatGPT behaves.
GPT-6.1 Terra Pricing and Availability
The new GPT-6.1 Terra is now accessible to Plus, Pro, Enterprise and Edu users in both ChatGPT Work and Codex. Developers are able to access it via the API under the model name gpt-6.1-terra.
Standard API costs in USD per million tokens are:
- Input: $1.00.
- Cached input: $0.05.
- Output: $5.00.
For companies that have a lot of software-based operations, there is a cheaper alternative, GPT-6.1 Terra. The proper choice relies on the intricacy of the work, the need for human inspection, and the permissions given to the agent.





