🔍 Read the full analysis: A Budget-Minded Guide To OpenAI’s GPT-6.1 Terra on ThorstenMeyerAI.com
Get smart everyday buys 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 per task at $0.055 on low effort and $0.40 at maximum effort, while finding that higher scores and lower costs are available from some competing models. Terra’s value depends on the task and effort setting, not just its listed token price.
OpenAI released GPT-6.1 Terra on October 26, 2026, positioning it between its smaller Luna model and larger Sol and Astra models, with API prices of $1 per million input tokens and $5 per million output tokens. Launch-day measurements from Artificial Analysis found that Terra’s cost per completed benchmark task varied from about $0.055 at low effort to $0.40 at maximum effort, making its appeal dependent on workload and settings rather than token price alone.
OpenAI says Terra is designed for communication and computer-based work, including high-volume tasks such as reading invoices and entering details into accounting software. The company lists the model in ChatGPT Work, Codex and its API as gpt-6.1-terra. It says the API rates are 50% lower than those for GPT-5.6 Terra and GPT-6.1 Sol, and that eligible ChatGPT Work and Codex subscribers can complete substantially more tasks within the same allowance than with Sol or Astra.
Artificial Analysis, an independent benchmarking firm, scored Terra at 48.1 on its Intelligence Index at maximum effort, at an average cost of $0.40 per task across its evaluation set. 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 figures describe that benchmark’s tasks and costing method; they are not a guarantee of performance or expense for a particular customer.
The six effort settings show a trade-off between score and cost. Artificial Analysis measured a score of 36.2 and $0.055 per task at low effort, compared with 48.1 and $0.40 at maximum effort. OpenAI’s launch materials also report improvements on selected tests: at medium effort, Terra scored 57.6% on Agents’ Last Exam, and at maximum effort it solved 50.3% of Terminal-Bench 4.0 tasks. Those are OpenAI-reported results, not independent measurements in the cited comparison.
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 Depends on the Task
For companies running large numbers of routine jobs, a model’s cost per finished task can be more useful than its token rate. A model with a cheaper per-token price may produce more tokens or need more work to complete a job; a model with a higher price may use fewer tokens or require fewer attempts. Artificial Analysis’s measurements give buyers a way to compare those costs on its particular set of tasks, while showing that Terra’s price advantage varies by effort setting.
The choice is not simply Terra versus a more expensive model. At maximum effort, the benchmark gave Terra a lower average task cost than Sol and Opus 5.5, but both scored higher. Artificial Analysis also listed Xiaomi’s open-weight MiMo-V2.6-Pro at a score of 46.3 and $0.13 per task, better on both measures than Terra at its highest settings. That comparison underlines why organizations should test their own workloads, including accuracy and review needs, before choosing a default model.
For individual users, lower rates may stretch a usage allowance, but they do not make Terra the right choice for every request. OpenAI says Sol and Astra remain better suited to its most complex coding tasks. The practical question is whether Terra can complete a specific job reliably at a lower total cost, not whether its headline token price is lower.
Popular right now
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
Samsung 42-Inch S90H OLED 4K Smart TV Bundle
Vitamix 5200 Professional-Grade Blender, 64 oz. Container, Black
As an affiliate, we earn on qualifying purchases.
How Terra Fits OpenAI’s Lineup
OpenAI describes Terra as a middle-tier model, above Luna and below Sol and Astra. Its launch materials emphasize speed, price and computer-use work, rather than claiming it leads the most capable models on every measure. The company says Terra completes tasks more than three times faster than Astra and Sol, but Artificial Analysis had not published speed measurements for Terra at the time of the source report.
Artificial Analysis’s Intelligence Index combines ten evaluations, including agentic office work, terminal tasks, science and hallucination testing. Its published average task cost reflects those evaluations and includes token use such as cached input and reasoning tokens. At maximum effort, the firm said Terra gained six index points over GPT-5.6 Terra at less than a third of the previous model’s average cost per task: $0.40 versus $1.40.
The measurements also reflect different sources and should not be blended. Artificial Analysis published its independent benchmark results on launch day. OpenAI separately reported its own computer-use, coding and factual-error results. For example, OpenAI says Terra’s error rate on an internal set of error-inducing prompts was 16.7% at low effort and 4.3% at maximum effort; those figures come from the company’s own test, not the independent index.
What the Benchmarks Cannot Establish
The published figures do not establish how Terra will perform across every customer’s software, data or workflow. Artificial Analysis’s average cost per task comes from its own evaluation set, and OpenAI’s results come from tests selected and reported by the company. The source material does not provide enough detail to independently verify every test condition or reproduce each result.
Several practical points remain unsettled in the available reporting: independent speed results were not yet published; real-world costs can differ with prompt length, retries, caching and the effort setting; and the source does not give the full benchmark results for all six effort levels. OpenAI also says Terra meets its Critical threshold in cybersecurity under its Preparedness Framework, but that is the company’s classification, not an independent assessment described in the material.
Benchmark scores also do not settle whether Terra is accurate enough for a particular business process. A low average cost can be offset if a task needs human correction or if errors carry substantial consequences. Buyers should treat the reported costs and scores as comparative measurements, not assurances about their own results.
Testing Terra in Real Workflows
Terra is available through ChatGPT Work, Codex and the API, so customers can begin comparing it with other models on their own jobs. A practical evaluation would track cost per successfully completed task, error rates, speed and the amount of human review required, while testing more than one effort setting. Routine, repeatable work may warrant low or medium effort; harder requests may need a more capable model or a higher setting.
Further independent measurements, particularly for speed and the other effort levels, would help clarify how Terra performs outside the reported maximum-effort comparison. For now, the confirmed development is the launch and its published pricing; the extent of savings for any individual user or business remains dependent on its workload and results.
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 priced at $0.05 per million tokens, according to the source material. Actual task costs depend on token use and model settings.
Is Terra OpenAI’s most capable model?
No. OpenAI places Terra between Luna and the larger Sol and Astra models. Artificial Analysis’s launch-day index also scored Terra below Sol and several other models at the settings reported.
Which Terra effort setting is cheapest?
Artificial Analysis measured the lowest average task cost at low effort: $0.055. The source also reports that non-reasoning mode cost nearly twice as much per task as low effort while scoring lower, so the cheapest option is not necessarily the setting with no reasoning.
Does a lower token price mean a lower bill?
Not necessarily. Total expense depends on the tokens used and whether the model completes the job successfully. The reported benchmark shows why buyers should compare cost per completed task alongside per-token rates, while recognizing that benchmark averages may not match their workloads.
Where can people use GPT-6.1 Terra?
OpenAI lists Terra in ChatGPT Work, Codex and the API, where the API model identifier is gpt-6.1-terra. Availability and account eligibility may depend on the product plan.
Source: ThorstenMeyerAI.com
