🔍 Read the full analysis: GPT-6.1 Terra: What You Gain By Choosing A Lower-Cost Model on ThorstenMeyerAI.com
Get hardware and tech essentials 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 found that Terra’s low- and medium-effort settings offer relatively strong benchmark performance for their measured task cost, but the model scores below several more capable competitors. The results are benchmark measurements, not guarantees of performance or savings for every workload.
OpenAI released GPT-6.1 Terra on October 26, a lower-cost model positioned between its smaller Luna model and larger Sol and Astra models. Its API price is $1 per million input tokens and $5 per million output tokens, half the listed price of its predecessor, GPT-5.6 Terra; independent testing found that Terra’s cheaper effort settings can deliver a relatively strong balance of benchmark score and task cost, though it does not match higher-scoring models.
OpenAI says Terra is designed for routine communication and computer-based work, including processing invoices. The model is available through the API as gpt-6.1-terra and to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. OpenAI says users’ allowances cover substantially more tasks than with Sol or Astra, but that statement is the company’s characterization rather than a guarantee of individual usage outcomes.
Artificial Analysis measured Terra’s Intelligence Index score at 48.1 in max-effort mode, with 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, Claude Opus 5.5 scored 57.6 at $5.98, and GPT-6 Luna scored 38.1 at $0.07. These are benchmark averages; costs and results for a particular user’s tasks may differ.
Terra offers six effort settings, trading greater reasoning effort and cost for higher measured scores. Artificial Analysis reported that low effort scored 36.2 for an average $0.055 per task, while non-reasoning scored 33.5 and cost $0.107. The source material says the six settings span scores from 33 to 48 and task costs from $0.05 to $0.40. At maximum effort, Terra costs about 7.4 times as much as low effort for 12 additional index points, according to the reported measurements.
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 Task Costs Change the Comparison
The launch highlights a distinction that matters to companies deploying AI at volume: token prices do not determine the final cost of completing a task. A model’s total bill also depends on how many input, output and reasoning tokens it uses, and whether it successfully completes the work. Artificial Analysis found Terra’s low and medium settings on its cost-efficiency frontier for the models and tasks it tested: no tested model was both cheaper and better at those points. That finding applies to the benchmark, not every use case.
The contrast with Luna illustrates the point. Luna’s listed token price is lower, but the source reports that Terra at low effort costs less per benchmark task because it generates fewer tokens. Organizations considering Terra should compare costs per completed task on their own workload, while tracking accuracy, rework and any human review required. A low benchmark cost is not evidence that every deployment will save money.
Terra also does not remove the need to select models by task. Artificial Analysis placed multiple systems above Terra’s max-effort score, while OpenAI says Sol and Astra remain better choices for its most complex coding work. The practical trade-off is lower cost for many routine tasks versus lower measured capability than larger or higher-scoring alternatives.
Products Worth Considering
LG UltraGear 32GX870A-B 32-inch 4K OLED Gaming Monitor
Apple MacBook Pro 14-inch with M5 Max Chip (18-Core CPU, 20-Core GPU), 48GB RAM, 2TB SSD, Space Black
LG UltraGear 32GX850A-B 32-Inch 4K UHD Glossy OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
Terra Between Luna and Sol
OpenAI describes Terra as a model for high-volume work and says its API rates are 50% lower than GPT-5.6 Terra’s listed $2 per million input tokens and $12 per million output tokens. Terra’s cached-input price is $0.05 per million tokens, according to the source material. The launch places the model in a tier between Luna and the company’s larger Sol and Astra offerings.
OpenAI’s published comparisons include computer-use, factual-error and coding tests. On Agents’ Last Exam, the company reports that 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 medium effort. OpenAI also reports a 60.0% score for Terra at max effort versus 63.2% for Claude Opus 5.5, with respective costs of $1.01 and $8.82 on that test. These are OpenAI-reported results, distinct from Artificial Analysis’s independent index.
OpenAI also reports that Terra completed tasks more than three times faster than Sol and Astra, and that its error rates on an internal set of error-inducing prompts were lower than GPT-5.6 Terra’s. Artificial Analysis had not published speed measurements for Terra when the cited evaluation was retrieved on October 26, 2026. Its separate Coding Agent Index gave Terra 48 points, compared with 54 for Sol and 31 for Luna.
What the Benchmarks Cannot Settle
Benchmark results do not establish how Terra will perform on a particular organization’s tasks. The available source does not provide independent speed measurements, and it does not quantify real-world savings after setup, review, retries or errors. OpenAI’s speed and internal error-rate figures are company-reported; they should not be treated as equivalent to the independent Intelligence Index results.
The supplied figures give full effort-setting details only for non-reasoning and low effort, alongside the overall score and cost ranges. They do not provide the complete values for the remaining settings, including the medium setting’s exact index score and task cost. The source also does not specify how the reported savings vary across industries or task types. Terra’s cybersecurity classification is described as OpenAI’s own assessment under its Preparedness Framework, not an external certification.
Testing Terra on Real Workloads
Terra is available now through the API, ChatGPT Work and Codex for the listed customer plans. The next useful measure for adopters is whether the model’s lower rates translate into lower costs on their own completed tasks, with acceptable accuracy and review requirements. Teams evaluating it can compare effort settings against the same representative workload and include token use, retries and human correction in their accounting.
Further independent measurements, especially speed results and tests across practical business tasks, may clarify how Terra compares beyond the current benchmark sets. Until then, the published results support a limited conclusion: Terra adds a cheaper option to OpenAI’s model lineup, while the best setting and model depend on the task. AI outputs can be wrong, and costs and performance can vary; benchmark comparisons are not guarantees of results.
Key Questions
What is GPT-6.1 Terra?
GPT-6.1 Terra is a new OpenAI model positioned between Luna and the larger Sol and Astra models. OpenAI describes it as suited to routine communication and computer-based tasks.
How much does Terra cost?
The listed API price is $1 per million input tokens and $5 per million output tokens. Cached input is listed at $0.05 per million tokens. Actual task costs depend on token use and the work being performed.
Is Terra cheaper than GPT-6 Luna?
Terra costs more per token than Luna at the listed rates, but Artificial Analysis reported that Terra at low effort had a lower average cost per benchmark task. The comparison reflects that evaluation set and should not be assumed to hold for every workload.
Does Terra outperform OpenAI’s larger models?
Not across the reported comparisons. Artificial Analysis scored Terra below Sol and Astra on the cited index, and OpenAI says Sol and Astra remain better for its most complex coding tasks. Terra is a lower-cost option, not a universal replacement.
Which Terra effort setting should users choose?
The source data indicates that higher effort can bring higher scores at higher cost, while low effort was among the most cost-efficient settings in Artificial Analysis’s tests. Users should test settings on their own tasks and account for accuracy, retries and review needs; the benchmark does not identify one best setting for every use.
Source: ThorstenMeyerAI.com
