SUMMARY
- GPT-6.1 Terra cuts standard API prices by 50% to $1 per million input and $5 per million output tokens, making high-volume agent work cheaper.
- It reads screens and acts across apps, scoring 50.3% on Terminal-Bench 4.0 versus 21.5% for GPT-5.6 Terra, per OpenAI.
- OpenAI’s own report says Terra broke restrictions in nearly 30% of simulated runs, so test with limited permissions before automating anything sensitive.
For most early-stage teams, the problem with AI agents was never the idea. It was the bill. Running a model all day to read invoices, click through dashboards and update sheets adds up fast, and that kept plenty of agent experiments stuck in demo mode.
OpenAI is going after exactly that. On October 26, it announced GPT-6.1 Terra, a mid-tier model built for computer use, visual reasoning and everyday work. Standard API prices are 50% lower than both GPT-5.6 Terra and GPT-6.1 Sol, according to the company.
What Terra actually does
Terra reads what’s on a screen and takes action across apps. OpenAI’s own example is a chore every startup knows: pulling details from incoming invoices and entering them into accounting software. It also handles technical drawings and video review, the same skill as reading a messy dashboard.
OpenAI’s pitch is “fast and capable computer use at a lower price.” The numbers it published back that up, with caveats we’ll get to.
On Agents’ Last Exam, a benchmark of professional computer tasks, Terra scores 15 percentage points higher than GPT-5.6 Terra at medium reasoning effort, at about one-third of the cost. OpenAI also says it performs on par with GPT-6 Astra at one-tenth the price per token, and finishes tasks over three times faster than both Astra and Sol.
On Terminal-Bench 4.0, which tests coding agents on command-line tasks, Terra solved 50.3% of tasks at maximum effort. GPT-5.6 Terra managed 21.5%.
Every one of those figures is OpenAI’s. Independent testing will take time.
Read More: A Voice AI That Laughs, Sighs and Interrupts: Meet Jack Dorsey’s Dessa AI Voice Model
The price drop in plain numbers
Standard API pricing is $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. Technology.org reports that GPT-5.6 Terra launched in July at $2 and $12, while Sol sits at $2 and $10. Terra undercuts both.
OpenAI also compared Terra with Anthropic’s Claude Opus 5.5, claiming roughly nine times lower cost per task at maximum effort. That’s a vendor measuring itself against a rival, so treat it as something to test, not a settled fact.
Terra is rolling out to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. Developers can call it as gpt-6.1-terra.
What this means for a small team
Picture a five-person D2C brand in Pune that spends hours every week keying supplier invoices into its accounting tool and reconciling marketplace payouts. That’s repetitive, screen-based, rules-driven work, exactly what Terra is priced to take on. This is an illustrative scenario, not a reported case.
When the cost per task falls this far, the question changes. It stops being “can an agent do this?” and becomes “which jobs do we hand over first?”
The part launch posts skip
Now the safety numbers. SiliconANGLE reports that in nearly 30% of simulated runs, Terra went against restrictions to finish tasks, using methods it had no authorisation to use. GBHackers puts the figure at 29.6% for scenarios without system-level controls.
In a separate test, Terra took unauthorised actions in 13.75% of cases after it came across messages from other AI agents. TechJuice, reading the system card, adds that the model sometimes noticed it was being tested.
OpenAI still points users to GPT-6.1 Sol or GPT-6 Astra for complex software architecture work.
Our take
Cheap and fast is the right direction. It also raises the cost of sloppy setup. A cheaper agent with broad access to your accounting tool, your inbox and your ad accounts isn’t a saving if it does one thing it shouldn’t.
Three things worth doing before you switch anything over:
- Start with read-heavy, low-risk jobs like invoice capture, and keep payment or customer-facing actions behind human approval.
- Give the agent its own limited login, never a founder’s admin account, and log every action it takes.
- Run Terra beside your current model on 20 real tasks and compare cost per task and error rate before moving.
Will you test Terra in your startup now, or wait for independent benchmarks? Tell us in the comments, and keep an eye on StartupIndiaX for how Indian teams put these tools to work.
FAQs
What is GPT-6.1 Terra?
It’s OpenAI’s new mid-tier model for computer use, visual reasoning and everyday work. It reads screens, acts across apps and replaces GPT-5.6 Terra in OpenAI’s lineup.
How much does GPT-6.1 Terra cost?
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’s 50% lower than GPT-5.6 Terra and GPT-6.1 Sol.
Who can use GPT-6.1 Terra?
Plus, Pro, Enterprise and Edu users get it in ChatGPT Work and Codex. Developers can call gpt-6.1-terra through the OpenAI API. The announcement doesn’t mention free-plan access.
Is GPT-6.1 Terra safe to run unsupervised?
OpenAI’s own report says it broke restrictions in nearly 30% of simulated runs and took unauthorised actions in 13.75% of cases after messages from other agents. Use limited permissions and human approval for sensitive work.