🔍 Read the full analysis: How To Decide Whether GPT-6.1 Terra Is Right For Your AI Workload on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
OpenAI released GPT-6.1 Terra on Oct. 26, 2026, positioning it as a lower-cost model for everyday work and computer-based tasks. Artificial Analysis found low- and medium-effort settings offer a strong measured score-to-cost balance, but Terra trails larger models on its Intelligence Index and is not the best fit for every task.
OpenAI released GPT-6.1 Terra on Oct. 26, a model positioned between its smaller Luna and larger Sol and Astra models, with API rates of $1 per million input tokens and $5 per million output tokens. Independent testing by Artificial Analysis found that Terra’s low- and medium-effort settings offer a favorable balance of benchmark score and measured task cost, while its highest-effort mode remains less capable than several larger models.
Artificial Analysis gave Terra a 48.1 Intelligence Index score at maximum effort and an average cost of $0.40 per task across the evaluations included in its index. In the same comparison, GPT-6.1 Sol scored 51.8 at $0.72 per task, while Claude Opus 5.5 scored 57.6 at $5.98. These are benchmark results, not guarantees about the cost or performance of a particular customer’s workload.
Terra has six reasoning-effort settings, ranging from non-reasoning to maximum. Artificial Analysis’s measurements show a curve from 33.5 points at non-reasoning to 48.1 at maximum, with the maximum setting costing about 7.4 times as much per task as low effort for a 12-point score gain. The source says low and medium settings sit on the evaluator’s cost-efficiency frontier; higher settings do not. Its supplied figures also show non-reasoning costs nearly twice as much per task as low effort while scoring lower.
OpenAI lists Terra in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, as well as through the API under the model name gpt-6.1-terra. OpenAI says its API prices are 50% below GPT-5.6 Terra’s, and says the same ChatGPT allowance can cover substantially more tasks than with Sol or Astra. Those availability and allowance descriptions come from OpenAI; the practical savings will depend on task volume, prompt and output length, and the selected effort setting.
Infographic 01 · The effort curve
One model, six settings, a 7-fold cost range.
GPT-6.1 Terra at each effort level, as measured by Artificial Analysis. Low and medium sit on its cost-efficiency frontier: no tested model is both cheaper and better.
Score · Intelligence Index
independent higher is better
Cost per finished task
note the last step shorter is cheaper
All six settings
| Effort | Index | Cost / task | Output tokens / task | Hallucination rate |
|---|---|---|---|---|
| Non-reasoning | Index 33.5 | Cost $0.107 | Tokens 4,737 | Hallucination 72% |
| Low | Index 36.2 | Cost $0.055 | Tokens 2,681 | Hallucination 46% |
| Medium | Index 41.3 | Cost $0.095 | Tokens 5,835 | Hallucination 44% |
| High | Index 43.9 | Cost $0.134 | Tokens 9,027 | Hallucination 43% |
| Xhigh | Index 45.4 | Cost $0.194 | Tokens 14,551 | Hallucination 42% |
| Max | Index 48.1 | Cost $0.403 | Tokens 35,096 | Hallucination 38% |
Hallucination rate in AA-Omniscience: how often the model answers anyway when it does not know. Lower is better.
The first steps up are cheap. The last one doubles the bill.Assessment: Thorsten Meyer
Sources: Artificial Analysis model pages (retrieved 26 Oct 2026); Artificial Analysis launch article (26 Oct 2026). Cost per task is the average across the index’s evaluation tasks.
Powered by Thorsten Meyer AI
Task Costs Shape Terra’s Best Uses
The launch highlights why token price alone can mislead when organizations compare AI models. A model with a cheaper rate per token may generate more tokens or need more work to complete a task. Artificial Analysis’s comparison accounts for measured spending across its own evaluation tasks, and its reported results place Terra’s low and medium settings favorably on that measure. It does not establish that Terra will be cheaper for every business process.
For high-volume, routine work—such as the invoice processing example OpenAI gives—lower task costs could make a capable model more economical to deploy. But the benchmark results also show a trade-off: Terra’s maximum setting costs more than its low setting, while its score remains below Sol, Astra and several competing models. Teams should test representative tasks and compare cost per successfully completed task, including error review and retries, rather than selecting a model from listed token rates alone.
There are also task-specific signs of improvement in OpenAI’s reported tests. On Terminal-Bench 4.0, OpenAI says Terra at maximum effort solved 50.3% of tasks, compared with 21.5% for the published GPT-5.6 Terra result. OpenAI also reports lower error rates on an internal set of prompts. These are company-reported results, and do not replace independent testing across a buyer’s own coding, data-entry or customer-service workload.
Products Worth Considering
Bose SoundLink Flex (2nd Gen) Portable Bluetooth Speaker
Amazon Kindle Paperwhite (12th Generation, 2024), 16 GB, 7-inch Display, Jade
Sonos Era 100 SL Compact, Microphone-Free Speaker – Black
As an affiliate, we earn on qualifying purchases.
Terra Between Luna and Sol
OpenAI describes Terra as a fast, efficient model for communication and computer-based tasks that fill a workday. Its example is software reading incoming invoices and entering their details into accounting systems. The company also says Terra completes tasks more than three times faster than Astra and Sol, but the supplied source notes that Artificial Analysis had not published speed measurements for Terra when its launch-day comparison was checked.
In Artificial Analysis’s index, Terra’s maximum-effort score falls below GPT-6.1 Sol, GPT-6 Astra, Claude Opus 5.5, Kimi K3.1 Pro and Gemini 4 Argon. It scores above GPT-6 Luna in that particular comparison. The source also identifies Xiaomi’s open-weight MiMo-V2.6-Pro at 46.3 points and $0.13 per task, a lower measured cost and score combination than Terra’s high and extra-high settings. Such rankings reflect the index’s selected evaluations and recorded costs, not every model capability or deployment condition.
Artificial Analysis says its Intelligence Index combines ten evaluations, including agentic office work, terminal tasks, science and hallucination tests, and tracks spending on those evaluations, including cached input and reasoning tokens. Its model-page figures were retrieved on Oct. 26, 2026. OpenAI separately reported Terra’s performance on tests including Agents’ Last Exam and Terminal-Bench 4.0; those results should be distinguished from the independent index.
Workload Results Still Need Testing
The supplied material does not provide a complete table of scores and costs for every one of Terra’s six effort settings, so the precise medium, high and extra-high figures cannot be independently restated here. Nor does it establish how the index’s average task cost maps to a particular organization’s usage. Real-world spending depends on task mix, token volume, caching, retries, human review and the chosen effort setting.
OpenAI’s claims about speed, internal factual-error rates, cybersecurity classification and ChatGPT allowance coverage are company statements in the source material. The source does not provide independent speed measurements, and it does not establish whether those internal results generalize to customer workloads. The durability of the benchmark rankings, availability across regions and any changes to pricing or access are also not specified.
Compare Terra on Real Tasks
Organizations considering Terra can start by testing low and medium effort on representative, repeatable tasks, then compare results with Luna, Sol or other available models. Record completion rate, error rate, latency and total cost, including the expense of retries and human checks. More demanding tasks can be tested at higher effort, but the launch-day measurements suggest buyers should confirm that the additional score is worth the added task cost.
Further independent testing, including speed measurements and a fuller comparison across all six settings, would help clarify where Terra fits. OpenAI’s own guidance says Sol and Astra remain better choices for its most complex coding tasks. No later benchmark release or timetable is specified in the source material, so the next confirmed step is workload-level evaluation by users, rather than a stated upcoming product milestone.
Key Questions
What is GPT-6.1 Terra?
GPT-6.1 Terra is a newly released OpenAI model positioned between Luna and the larger Sol and Astra models. OpenAI describes it as suited to everyday communication and computer-based work.
How much does GPT-6.1 Terra cost?
The listed API rate is $1 per million input tokens and $5 per million output tokens, with cached input listed at $0.05 per million tokens in the source material. Those rates are separate from Artificial Analysis’s average cost-per-task estimates.
Which Terra effort setting should teams try first?
Artificial Analysis’s reported results place low and medium effort on its cost-efficiency frontier. Teams should test those settings first on their own tasks, since benchmark averages do not predict every workload’s cost or accuracy.
Is Terra better than GPT-6.1 Sol or Claude Opus 5.5?
Not on Artificial Analysis’s reported maximum-effort Intelligence Index comparison: Sol scored 51.8 and Opus 5.5 scored 57.6, compared with Terra’s 48.1. Terra’s measured average task cost was lower, so the choice depends on the required capability and budget.
Does Terra’s lower token price guarantee lower operating costs?
No. Token rates do not account by themselves for how many tokens a task uses or whether it needs retries and human review. Compare total cost per successfully completed task on representative work before choosing a model.
Source: ThorstenMeyerAI.com
