🔍 Read the full analysis: GPT-6.1 Terra Might Be The Budget-Friendly Way To Get The Job Done on ThorstenMeyerAI.com
Get privacy and security gear 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 API use at $1 per million input tokens and $5 per million output tokens. Artificial Analysis measured its cost and performance across tasks, finding that low and medium effort settings sit on its cost-efficiency frontier; Terra does not lead the benchmark, and higher effort costs more without matching the cheaper settings’ efficiency.
OpenAI released GPT-6.1 Terra on October 26, positioning it between the smaller Luna and larger Sol and Astra models and setting API prices at $1 per million input tokens and $5 per million output tokens. Independent testing by Artificial Analysis found that Terra’s low and medium reasoning settings offer a more favorable balance of task scores and cost than its higher settings, though Terra does not top the overall benchmark.
OpenAI describes Terra as a fast, efficient model for communication and computer-based work, including processing invoices and entering their details into accounting software. The company says it is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, as well as through the API under the model name gpt-6.1-terra. Its API rates are half those listed for GPT-5.6 Terra and GPT-6.1 Sol; cached input is priced at $0.05 per million tokens.
Artificial Analysis’s launch-day evaluation gave Terra a 48.1 Intelligence Index score at maximum effort, at an average cost of $0.40 per task across the benchmark’s evaluation set. For comparison, it listed GPT-6.1 Sol at 51.8 points and $0.72 per task, and Claude Opus 5.5 at 57.6 points and $5.98 per task. The scores and costs describe different trade-offs: Terra was cheaper in this test, while both comparison models scored higher.
The reported cost curve varies by reasoning setting: Terra scored from 33.5 at non-reasoning to 48.1 at maximum effort, while average task cost ranged from about $0.05 at low effort to $0.40 at maximum. Artificial Analysis identifies low and medium effort as settings on its cost-efficiency frontier. That finding applies to its benchmark and pricing assumptions; it does not establish that Terra is the least expensive or best-performing option for every real-world task.
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
Why Per-Task Costs Matter
For organizations running high volumes of routine work, the relevant expense may be the cost to complete a task rather than the price of each token. A lower token rate alone does not settle the comparison: models can use different numbers of input, output and reasoning tokens to reach an answer. Artificial Analysis’s figures account for the costs it observed across its evaluation tasks, making its per-task comparison more informative than token rates alone for that particular workload.
Terra’s results suggest a potential middle option for users who need more capability than a small model provides but cannot justify using a more expensive model on every request. The benchmark also shows a limit to that case: higher effort raises both score and cost, and Terra’s maximum-effort score remained below several larger models. Companies considering it would need to test their own tasks, error rates and actual usage before estimating savings.
OpenAI also reports gains on selected internal and task-specific evaluations. On Agents’ Last Exam, the company says Terra at medium effort scored 57.6% and cost $0.29 per task, compared with 42.6% and $0.84 for GPT-5.6 Terra at the same effort. OpenAI says Terra’s maximum-effort score on Terminal-Bench 4.0 was 50.3%. These are company-reported results, not independent confirmations of performance across customers’ workloads.
Products Worth Considering
Jackery Explorer 1000 v2 Portable Power Station
Bose SoundLink Flex (2nd Gen) Portable Bluetooth Speaker
Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation
As an affiliate, we earn on qualifying purchases.
How Terra Fits OpenAI’s Lineup
OpenAI positions Terra between GPT-6 Luna and the larger Sol and Astra models. The source material describes Luna as substantially cheaper per token, but Artificial Analysis’s figures show why token price and task price can diverge: Luna’s maximum setting cost $0.07 per task in its evaluation, while Terra’s low-effort setting cost $0.055. Those are different settings and benchmark averages, not a universal comparison; the results depend on the tasks and token use measured.
Artificial Analysis says its Intelligence Index combines ten evaluations, covering areas such as agentic office work, terminal tasks and science. Its October 26 model-page figures place Terra below Claude Opus 5.5, Kimi K3.1 Pro, GPT-6 Astra, Gemini 4 Argon and GPT-6.1 Sol on score. It also lists the open-weight MiMo-V2.6-Pro at 46.3 points and $0.13 per task. This indicates Terra enters a field with other lower-cost options rather than an uncontested price category.
OpenAI’s launch material further reports that Terra is more than three times faster than Sol and Astra, and that its safeguards carry over from GPT-6 Astra. The company classifies Terra as meeting the “Critical” threshold in cybersecurity under its Preparedness Framework. These are OpenAI’s own statements; the supplied material does not include independent speed measurements or an external assessment of that security classification.
Limits of the Benchmark Evidence
Artificial Analysis’s results are averages across its selected evaluation tasks. The supplied material does not establish how those figures translate to particular business workflows, nor whether a customer will see the same task costs, quality or token use. The benchmark’s cost-efficiency frontier is specific to the models and settings it tested and the prices in effect on October 26, 2026.
Some performance figures come from OpenAI’s own launch-page charts, including results for computer use, coding, speed and factual errors. The source material does not provide independent replication of those claims. Artificial Analysis had not published speed measurements for Terra at the time its evaluation was checked. OpenAI’s cybersecurity classification is also its own assessment, not an outside finding in the material provided.
The source excerpt gives the score and cost range for Terra’s effort settings but does not supply the complete row-by-row figures for every setting. It is consequently not possible here to verify each intermediate setting’s cost, score or hallucination rate from the provided details. Terra’s longer-term pricing, availability changes and performance under subsequent independent tests also remain unsettled.
What Users Should Test Next
Users can access Terra through the listed ChatGPT Work and Codex plans or the API, subject to their account and product availability. For a practical comparison, teams can start with low or medium effort, then evaluate representative tasks for completion rate, factual errors, response time and total cost. The benchmark suggests those settings merit testing; it does not establish that they will be right for every workload.
For work that requires more reasoning, users can compare Terra at higher effort with Sol, Astra or other models, weighing any score gains against added expense. OpenAI’s own material says Sol and Astra remain preferable for its most complex coding tasks. Further independent speed measurements and evaluations on customer-relevant tasks would help clarify how far Terra’s launch-day results carry beyond the published benchmarks.
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. Cached input is listed at $0.05 per million tokens. Artificial Analysis measured an average cost of $0.40 per task at maximum effort across its evaluation set; actual costs depend on task and usage.
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 GPT-6 Astra at 52.7 in the figures provided. Terra’s launch case is its price and task-cost profile, not the highest score in that comparison.
Which Terra reasoning setting appears most cost-efficient?
Artificial Analysis says low and medium effort sit on its cost-efficiency frontier. That finding reflects its benchmark tasks and listed prices; users should test their own work before choosing a setting.
Are Terra’s reported performance gains independently verified?
Some results come from Artificial Analysis, which published an independent evaluation on launch day. Other claims about computer use, coding, factual errors, speed and cybersecurity come from OpenAI’s own launch material. The supplied source does not include independent confirmation of those company-reported measurements.
Source: ThorstenMeyerAI.com
