🔍 Read the full analysis: GPT-6.1 Terra Isn’t OpenAI’s Smartest Model. It May Be The Cheapest Way To Get The Job Done on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
OpenAI released GPT-6.1 Terra on October 26, 2026, pricing it at $1 per million input tokens and $5 per million output tokens. Artificial Analysis found that low- and medium-effort settings deliver comparatively low costs per benchmark task, while Terra’s highest-effort score remains below several competing models. Benchmark results and launch claims do not establish how it will perform on every customer’s workload.
OpenAI released GPT-6.1 Terra on October 26, adding a lower-priced model between its Luna and Sol/Astra offerings. The model costs $1 per million input tokens and $5 per million output tokens; independent benchmarking by Artificial Analysis found its low- and medium-effort settings on the firm’s cost-efficiency frontier, while its highest-effort score remains below several more capable models.
OpenAI describes Terra as a fast, efficient model for communication and computer-based work, including processing invoices and entering information into accounting software. The company says the model is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, as well as through its API under the model name gpt-6.1-terra. Its listed API prices are half those of GPT-5.6 Terra: $1 per million input tokens and $5 per million output tokens, with cached input priced at $0.05 per million tokens.
Artificial Analysis reported an Intelligence Index score of 48.1 for Terra at maximum effort, at an average cost of $0.40 per task across the index’s evaluations. That score was below GPT-6.1 Sol’s 51.8 and Claude Opus 5.5’s 57.6, but above GPT-6 Luna’s 38.1. The firm said Terra’s low and medium settings sit on its cost-efficiency frontier: among the models and settings it tested, no alternative was both less expensive per task and higher-scoring. Its published results put the six effort settings between 33.5 and 48.1 points, and between $0.055 and $0.40 per task.
The launch-day comparison also shows why token rates alone can mislead. Artificial Analysis measured Luna at $0.07 per task at maximum effort, versus $0.055 for Terra at low effort, even though Luna’s listed token price is lower. Terra used fewer output tokens in that comparison. The benchmark firm’s figures reflect its evaluation tasks and accounting method, not a guaranteed price for a customer’s own work. OpenAI separately reported improved results on selected computer-use, coding and factual-error tests; those are company-reported measurements, distinct from Artificial Analysis’s evaluation.
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
The Cost of Completing Work
For businesses using AI in repeated workflows, cost per completed task may be more useful than the price of each token. A model with a low token rate can still spend more if it produces longer outputs or needs more reasoning. Terra’s benchmark results suggest that lower-effort settings may offer a way to reduce spending on routine work, provided they meet the accuracy and reliability requirements of the task.
The trade-off is capability. Terra’s maximum-effort score is below several higher-scoring models, and Artificial Analysis found its highest effort setting cost more than seven times as much per task as its low setting for a 12-point gain on the index. OpenAI also says Sol and Astra remain better choices for its most complex coding tasks. The results point toward selecting a model and effort level by task, rather than assuming one model is the best choice for every request.
These findings matter most to organizations running high volumes of repeatable work, where small differences in cost per task can add up. They do not establish that Terra will be cheaper or more accurate across all workloads: actual spending depends on task design, input and output length, retries, and the model setting used.
Products Worth Considering
Samsung 32-Inch Odyssey OLED G8 (G80SH) 4K Gaming Monitor
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
Samsung 27-Inch Odyssey QD-OLED G8 4K 240Hz Gaming Monitor
As an affiliate, we earn on qualifying purchases.
How Terra Fits OpenAI’s Lineup
OpenAI places Terra between Luna and Sol/Astra in its current model family. The source report describes Luna as the smaller, lower-cost option and Sol and Astra as higher-capability choices. Terra’s lower published token prices compared with GPT-5.6 Terra and GPT-6.1 Sol are part of its pitch for work that is frequent and structured, rather than the hardest tasks.
Artificial Analysis’s Intelligence Index combines ten evaluations, including agentic office work, terminal tasks, science and hallucination tests, and reports average task costs alongside scores. Its launch-day figures rank Claude Opus 5.5, Kimi K3.1 Pro and GPT-6 Astra above Terra on the reported maximum-effort score. The index also includes cheaper alternatives, such as open-weight MiMo-V2.6-Pro, which scored 46.3 at an average $0.13 per task. These are benchmark comparisons, not a universal ranking for every use case.
OpenAI’s own launch materials cite other tests. The company reported that Terra at medium effort scored 57.6% on Agents’ Last Exam, and said its maximum-effort result on Terminal-Bench 4.0 was 50.3%. OpenAI also reported lower factual-error rates than for GPT-5.6 Terra on its internal prompt set. These figures should be read as OpenAI’s stated results; the available source does not provide independent confirmation of those specific tests.
What the Benchmarks Cannot Show
The available measurements do not establish how Terra will perform across real customer workloads, which can differ from the index’s test tasks. Artificial Analysis’s average cost per task is a benchmark-specific figure; it is not a published price for a defined commercial job. The supplied report also does not include the complete results for every effort setting, so comparisons beyond the stated figures should not be inferred.
OpenAI’s claims about speed, computer use, coding, factual errors and cybersecurity are company-reported. The source says Artificial Analysis had not published speed measurements for Terra when the report was checked. OpenAI’s statement that Terra meets the “Critical” threshold in cybersecurity is its own Preparedness Framework classification, not an independent assessment presented in the material.
It is also not clear from the available information how the model’s performance will vary by language, industry, tool setup or task complexity, or whether the same results hold under different API usage patterns. Buyers will need to test the model against their own requirements, including the cost of errors and human review.
Testing Terra in Real Workflows
Terra is available now through the channels OpenAI listed, so the next evidence will come from customer deployments and further independent testing. Organizations considering it can compare low and medium effort against their current model on representative tasks, tracking completion rates, output quality, latency and total cost rather than token rates alone.
Further benchmark results could clarify how Terra’s six effort settings compare on speed and task quality, and whether its cost advantage persists outside the evaluations published on launch day. OpenAI’s guidance points users toward Sol or Astra for its most complex coding work; the appropriate choice for other demanding tasks remains dependent on testing. For now, the confirmed development is a new, lower-priced model and an early set of benchmarks—not proof that Terra is the best-value option for every user.
Key Questions
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. Actual task costs vary with usage and the work being performed.
Is Terra OpenAI’s highest-scoring model?
No. Artificial Analysis scored Terra at 48.1 at maximum effort, below GPT-6.1 Sol at 51.8 and several other models in its comparison. OpenAI also says Sol and Astra are better suited to its most complex coding tasks.
Why compare cost per task instead of token price?
Token price does not account for how many tokens a model uses to finish a job. Artificial Analysis’s benchmark found Terra at low effort averaged $0.055 per task, while Luna at maximum effort averaged $0.07, despite Luna’s lower listed token rate.
Where can people use GPT-6.1 Terra?
OpenAI says Terra is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and through the API as gpt-6.1-terra.
Are Terra’s reported benchmark results independently verified?
Artificial Analysis published its own Intelligence Index evaluation. Other results in the source, including speed, selected computer-use and coding scores, and factual-error rates, are attributed to OpenAI and should be treated as company-reported measurements.
Source: ThorstenMeyerAI.com
