🔍 Read the full analysis: GPT-6.1 Terra Isn’t The Smartest—Could It Still Be The Savvier Choice? on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
OpenAI released GPT-6.1 Terra on October 26, 2026, positioning it as a lower-priced model for routine work rather than its highest-performing option. Artificial Analysis measured Terra at $0.40 per task and an Intelligence Index score of 48.1 at maximum effort; those figures suggest a potential cost advantage for some workloads, not a general performance lead.
OpenAI released GPT-6.1 Terra on October 26, introducing a model priced below its larger GPT-6.1 counterparts and aimed at routine communication and computer-based work. In Artificial Analysis’s launch-day evaluation, Terra scored 48.1 on the Intelligence Index at maximum effort and averaged $0.40 per task: cheaper than GPT-6.1 Sol and several other high-scoring models in that test, but not the top performer.
OpenAI lists Terra’s API price at $1 per million input tokens and $5 per million output tokens, compared with $2 and $12 for GPT-5.6 Terra. Cached input is listed at $0.05 per million tokens. The company says the model is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, as well as through the API under the name gpt-6.1-terra.
Artificial Analysis’s figures put Terra at 48.1 points and $0.40 per task at maximum effort. 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 averages across the benchmark’s evaluation tasks, not a prediction of what every user will pay. The evaluator’s index combines ten tests, including office work, terminal tasks, science and hallucination.
OpenAI’s own launch-page results describe specific tests rather than an independent overall ranking. The company reports that Terra at medium effort scored 57.6% on Agents’ Last Exam for $0.29 per task, compared with GPT-5.6 Terra’s 42.6% and $0.84. On Terminal-Bench 4.0, OpenAI reports a 50.3% solve rate at maximum effort, versus 21.5% for GPT-5.6 Terra. OpenAI also says Terra completes tasks more than three times faster than Sol and Astra; Artificial Analysis had not published speed measurements for Terra when the source report checked.
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
When Lower Cost Can Matter
The launch puts attention on a practical trade-off: model price per token is not the same as cost per finished task. A model that uses fewer tokens may complete a particular job for less, even if its listed token rate is higher than a smaller model’s. Artificial Analysis’s results support that distinction for its tested workloads: its maximum-effort Terra average was $0.40 per task, compared with $0.07 for GPT-6 Luna, while Terra scored ten points higher on the index. The task-cost figures reflect different output and reasoning-token use across the evaluation.
That could matter to businesses handling repeated, bounded jobs such as invoice reading and data entry, an example OpenAI gives for Terra. Lower average task costs can affect how many jobs fit within a budget, but the benchmark does not establish that Terra is the cheapest or best choice for every workflow. Results will depend on task difficulty, prompt design, error handling and the amount of human review required.
For individual users, Terra offers a possible middle option between a small model and larger, more expensive systems. It does not remove the need to choose by task: OpenAI says Sol and Astra remain preferable for its most complex coding work, and the benchmark ranks several models above Terra on overall score.
Products Worth Considering
Logitech MX Mechanical Full-Size Wireless Mechanical Keyboard, Tactile Quiet Switches, Backlit Keys, Bluetooth, USB-C, Multi-Device, Graphit
Dell XPS 13 9350 Laptop, 13.4 FHD+ Display, Intel Ultra 7 256V, 16GB DDR5, 1TB SSD
Apple iPad Air 13-inch (M4), 128GB, Wi-Fi 7, Space Gray
As an affiliate, we earn on qualifying purchases.
Terra Between Luna and Sol
OpenAI presents Terra as part of a model range, with Luna positioned as smaller and Sol and Astra as larger options. The company describes Terra as a “fast, efficient model” for communication and computer-based tasks that fill the workday. Its launch materials say the same ChatGPT Work allowance can cover substantially more Terra tasks than Sol or Astra tasks; that is OpenAI’s product claim, and the source provides no allowance comparison in absolute task counts.
Artificial Analysis’s results give the choice more detail. Terra’s maximum-effort score of 48.1 falls below Sol’s 51.8 and Astra’s 52.7, as well as Claude Opus 5.5’s 57.6. The report also identifies cheaper alternatives within its test: open-weight MiMo-V2.6-Pro scored 46.3 at $0.13 per task. Those rankings apply to the evaluator’s index and its listed prices, retrieved on October 26, 2026; they do not establish a universal ordering for all uses.
The source report says Artificial Analysis measured six Terra effort settings, with scores ranging from 33 to 48 and task costs from $0.05 to $0.40. Its summary says the low and medium settings sit on the cost-efficiency frontier in the tested set, meaning no tested model was both cheaper and better at those points. The supplied material does not include the complete setting-by-setting table, so the precise figures for every level cannot be independently detailed here.
Benchmark Limits and Open Questions
The available results do not show how Terra will perform across every customer’s software, data, prompts or review process. Benchmark averages are not guarantees of cost, accuracy or speed for an individual task. Artificial Analysis’s scores and prices describe its evaluation setup, while OpenAI’s results come from the company’s own tests and should be read as company-reported findings.
Some claims remain unverified by the independent measurements cited in the source report. OpenAI says Terra is more than three times faster than Sol and Astra, but Artificial Analysis had not published speed results for Terra at the time of checking. OpenAI also says the model meets the “Critical” threshold in cybersecurity under its Preparedness Framework; the cited material identifies this as the company’s classification, not an outside finding.
The source notes that Terra guessed on 38% of questions it could not answer in Artificial Analysis’s hallucination test, down from 88% for GPT-5.6 Terra. That comparison is limited to that test and its unanswered questions; it does not mean Terra’s overall hallucination rate is 38%, nor establish how often it will make errors in ordinary use.
How Terra Performs in Use
The next useful evidence will come from independent measurements of speed, broader testing across real workplace tasks and clearer comparisons of total costs at each effort setting. Businesses considering Terra can compare completed-task costs on their own workloads, including verification and correction time, rather than relying only on token prices or a benchmark average.
OpenAI’s launch makes Terra available through ChatGPT Work, Codex and the API, so users can begin testing it within those products. Whether it becomes a savvier choice will depend on whether its lower listed rates and benchmark task costs hold up for a given job—and whether its score is adequate for that job’s accuracy requirements.
Key Questions
Is GPT-6.1 Terra OpenAI’s highest-scoring model?
No. In Artificial Analysis’s launch-day results, Terra scored 48.1 at maximum effort, below GPT-6.1 Sol at 51.8 and GPT-6 Astra at 52.7. The benchmark also placed other models above Terra.
How much does GPT-6.1 Terra cost?
OpenAI lists API pricing at $1 per million input tokens and $5 per million output tokens, with cached input at $0.05 per million tokens. Artificial Analysis reported an average of $0.40 per task at maximum effort in its evaluation; that is a benchmark average, not a fixed charge for every task.
Why compare cost per task rather than token prices?
A task’s total cost depends on how many input, output and reasoning tokens a model uses. Cost per completed task can therefore differ from what token rates alone suggest, though actual costs vary by workload.
What work is Terra intended for?
OpenAI describes Terra as suited to communication and computer-based work, including high-volume tasks such as reading invoices and entering information into accounting software. The company says Sol and Astra remain better choices for its most complex coding tasks.
Are Terra’s speed and safety claims independently confirmed?
Not by the cited launch-day comparison. OpenAI says Terra is more than three times faster than Sol and Astra and describes its cybersecurity classification under its own Preparedness Framework. Artificial Analysis had not published Terra speed measurements when the source report checked.
Source: ThorstenMeyerAI.com
