GPT-6.1 Terra lowers OpenAI prices without compromising performance
OpenAI launches GPT-6.1 Terra, a mid-range model priced at half the cost of GPT-5.6 Terra. According to published tests, it improves cost-performance ratio and significantly reduces hallucinations.
OpenAI launched GPT-6.1 Terra on October 26, 2026, a mid-range model marketed as more cost-effective than GPT-5.6 Terra and designed for office tasks as well as software operation. Its appeal is straightforward: at the API scale, the price drops significantly while performance metrics remain strong. The model is already available in ChatGPT Work, Codex, and via the API for Plus, Pro, Enterprise, and Edu plans.
A model designed for everyday work, not just demos
OpenAI presents Terra as a model tailored for drafting proposals, interpreting technical plans, or filling out forms. The company also highlights its ability to combine rapid visual reasoning with precise computer control, placing it in the category of so-called agent models, meaning capable of chaining actions rather than just answering questions. In professional daily use, this matters as much as raw scores. Users don’t pay to admire a curve; they want the system to complete the task.
On pricing, OpenAI announces $1 per million input tokens and $5 per million output tokens, compared to $2 and $10 for Claude Sonnet 5.5, and $2 and $12 for GPT-5.6 Terra. Cached inputs drop to $0.05 per million tokens, a quarter of the previous rate. This positioning places Terra as a direct competitor to Anthropic’s Sonnet range, with a lower cost for a quality level the publisher now considers close to its largest models.
What tests say about cost-performance ratio
Artificial Analysis evaluations align with this. At maximum reasoning effort, GPT-6.1 Terra gains 6 points on the Intelligence Index compared to GPT-5.6 Terra, while costing less than a third per task: $0.40 versus $1.40. At lower effort levels, the gap widens further, with gains of 7 to 13 points. The model thus sits on what analysts call the Pareto frontier, the zone where it becomes difficult to improve both price and performance simultaneously.
On the Intelligence Index, Terra scores 48 points, between GPT-6.1 Sol at 52 and GPT-5.6 Terra at 42. In the Coding Agent Index v2.0, it reaches 48, six points less than GPT-6.1 Sol but 17 points more than GPT-6 Luna. A similar progression appears on code: in Terminal-Bench 4.0, the model solves 50.3% of tasks at maximum effort, versus 21.5% for GPT-5.6 Terra. This is a clear leap, not just a press release tweak.
Fewer hallucinations and better software agent performance
Hallucination reduction also stands out. On the private AA-Omniscience benchmark, the model “guesses” only 38% of the time when it doesn’t know the answer, versus 88% for GPT-5.6 Terra. Accuracy improves by two points. For business use, this kind of decrease is far from trivial: fewer inventions mean less manual verification and less time wasted correcting the tool.
OpenAI also emphasizes software control features. On the Agents’ Last Exam, Terra matches GPT-6 Astra’s performance for a tenth of the token cost. Compared to Claude Opus 5.5, the model is over three times more token-efficient at a quarter of the price, which according to the company equates to about nine times lower cost per task. It also finishes faster than GPT-6 Astra and GPT-6.1 Sol, again according to OpenAI.
Security and controlled use: the “Critical” level
Terra meets the “Critical” cybersecurity threshold defined by OpenAI’s Preparedness Framework and inherits GPT-6 Astra’s safeguards. According to the system card, the model resists direct attacks on instruction hierarchy as well as GPT-6 Astra and GPT-6.1 Sol, with a 99.99% success rate on these tests. For indirect injections, where malicious instructions are hidden in third-party content, the defense rate reaches 98.8%, versus 96.2% for GPT-5.6 Terra.
The overall picture is clear: OpenAI is not just pushing a cheaper model but one that tightens costs without sacrificing the most profitable use cases. One important nuance remains: for the most demanding coding tasks, the company still recommends GPT-6.1 Sol and GPT-6 Astra. In short, Terra moves upmarket but does not yet claim to do everything. And that is probably more honest than a slogan with glitter.
Key points
- Launched on October 26, 2026 by OpenAI.
- GPT-6.1 Terra costs half as much as GPT-5.6 Terra.
- $1 per million input tokens, $5 per million output tokens.
- Artificial Analysis gives it 48 points on the Intelligence Index.
- AA-Omniscience: 38% guessed responses, versus 88%.
- OpenAI ranks it at the “Critical” cybersecurity level.
By the numbers
- $0.40 — cost per task at maximum effort, according to Artificial Analysis.
- $1.40 — cost per task for GPT-5.6 Terra, according to Artificial Analysis.
- 50.3% — resolution rate on Terminal-Bench 4.0, at maximum effort.
- 98.8% — defense rate against indirect injections, according to the system card.
- 99.99% — resistance to direct attacks on instruction hierarchy.