🔍 Read the full analysis: GPT-6.1 Terra: Less Focus On Smarts, More On Getting The Job Done on ThorstenMeyerAI.com
Get business pricing on tech for your team
- 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-cost model for routine workplace and computer-based tasks rather than its highest-scoring model. Artificial Analysis measured a maximum-effort score of 48.1 at an average $0.40 per task; its reported figures put Terra’s low and medium settings on the cost-efficiency frontier. The results are benchmark measurements, not a guarantee of performance on every user’s work.
OpenAI released GPT-6.1 Terra on October 26, presenting it as a lower-cost model for communication and computer-based work rather than a replacement for its top-performing models. The API price is $1 per million input tokens and $5 per million output tokens, half the listed rates for GPT-5.6 Terra; independent benchmark firm Artificial Analysis measured Terra at an average $0.40 per task at maximum effort.
Terra 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. OpenAI describes it as a model for work such as processing incoming invoices and entering data into accounting software. The company says its ChatGPT allowance can cover substantially more tasks with Terra than with Sol or Astra, but did not provide a quantified allowance comparison in the supplied launch material.
Artificial Analysis’s Intelligence Index gives Terra 48.1 points at maximum effort, compared with 51.8 for GPT-6.1 Sol and 52.7 for GPT-6 Astra. The firm reports average benchmark-task costs of $0.40 for Terra, $0.72 for Sol and $5.98 for Claude Opus 5.5. Those results place Terra below several models in the index on score, while its task cost is lower than Sol’s and Opus 5.5’s.
Terra has six effort settings, from non-reasoning to maximum. Artificial Analysis reports that the low setting costs an average $0.055 per task and scores 36.2; medium costs $0.?? [sic]. The source material says the six settings span scores from 33 to 48 and task costs from $0.05 to $0.40, and that low and medium are on the firm’s cost-efficiency frontier. At maximum effort, the source reports 12 more points than low for 7.4 times the cost. The figures are averages across benchmark tasks, not prices for a fixed amount of real-world work.
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 Cost Changes the Comparison
Terra’s launch highlights a practical distinction for businesses choosing models: token price is not the same as the cost of completing a task. A cheaper model can use more tokens or require more work to reach an answer, while a higher-priced model may finish with fewer. Artificial Analysis says Terra at low effort can cost less per benchmark task than GPT-6 Luna, despite Luna’s lower listed per-token price, because Terra produces fewer tokens in those tests.
For organizations that run large volumes of routine work, the relevant budget question may be how much each completed workflow costs, alongside accuracy and review requirements. The benchmark results suggest starting with low or medium effort for suitable tasks, then raising effort or routing difficult work to a larger model when needed. They do not establish that Terra will be cheaper or better for every business process: task mix, input length, error costs and human oversight can change the economics.
The trade-off is visible in the scores. Terra’s maximum setting is cheaper per task than several higher-scoring competitors in Artificial Analysis’s index, but it does not match their measured performance. The source also reports that Xiaomi’s open-weight MiMo-V2.6-Pro scored 46.3 for $0.13 per task, a lower cost and higher score than Terra’s high and extra-high settings in the comparison. Buyers therefore have more than a simple choice between a premium model and Terra.
Products Worth Considering
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
Samsung 32-Inch Odyssey OLED G8 (G80SH) 4K Gaming Monitor
As an affiliate, we earn on qualifying purchases.
Terra Within OpenAI’s Model Lineup
OpenAI places Terra between its smaller Luna model and larger Sol and Astra models. The new model’s listed API rates are 50% lower than GPT-5.6 Terra’s, according to the launch material. The release is framed around routine workplace throughput, not a claim that Terra leads general-purpose intelligence rankings.
OpenAI’s own published evaluations make a separate case for improvements over its predecessor. On Agents’ Last Exam at medium effort, the company reports a 57.6% score for Terra, 15 percentage points above GPT-5.6 Terra, at $0.29 per task versus $0.84. OpenAI also reports a 50.3% result on Terminal-Bench 4.0 at maximum effort, compared with 21.5% for GPT-5.6 Terra. These are company-reported benchmark results; they should not be conflated with Artificial Analysis’s independent index.
On OpenAI’s internal error-inducing prompts, the company says Terra’s error rate declined from 53.6% to 16.7% at low effort and from 22.3% to 4.3% at maximum effort, relative to GPT-5.6 Terra. Artificial Analysis separately reports that Terra guessed on 38% of questions it could not answer in its hallucination test, down from 88% for GPT-5.6 Terra. The tests use different measures and should be read as separate findings.
Limits of the Benchmark Evidence
The reported figures do not show how Terra performs across every real workplace workflow, nor do they establish a universal cost per completed task. Artificial Analysis’s task-cost figure is an average across its index evaluations; actual API bills will vary with prompts, outputs, cached inputs, reasoning effort and repeated attempts. The source material does not provide the full results for every effort setting, so a complete setting-by-setting comparison cannot be reproduced here.
OpenAI says Terra completes tasks more than three times faster than Sol and Astra, but Artificial Analysis had not published speed measurements when the source checked on launch day. The speed claim is consequently attributable to OpenAI, not independently verified in the cited evaluation. OpenAI also says Terra meets the Critical threshold in its cybersecurity Preparedness Framework; that classification is the company’s own assessment.
It remains unclear how the model will perform on specific customer workloads, how its reliability compares after deployment, and whether independent speed and broader real-world evaluations will support OpenAI’s claims. The supplied material also does not give a full breakdown of all six effort settings’ costs, scores and error rates.
How Buyers Can Test Terra
Terra is available now through ChatGPT Work, Codex and the API, so users and companies can test it against their own tasks. A practical evaluation would compare cost per successfully completed workflow, not token rates alone, while tracking accuracy, latency, retries and the time needed for human review.
For now, the launch data supports treating Terra as a lower-priced option for routine tasks and checking whether low or medium effort is sufficient before assigning harder work to a more capable model. Further independent speed measurements, expanded benchmark results and customer experience data could clarify how well its test performance carries over to production.
Key Questions
What is GPT-6.1 Terra?
GPT-6.1 Terra is OpenAI’s new model for communication and computer-based tasks. It sits between Luna and the larger Sol and Astra models in OpenAI’s lineup.
How much does Terra cost?
OpenAI’s listed API price is $1 per million input tokens and $5 per million output tokens, with cached input at $0.05 per million tokens. Artificial Analysis reports an average cost of $0.40 per benchmark task at maximum effort; that is a benchmark average, not a fixed charge for every task.
Is Terra OpenAI’s highest-scoring model?
No. Artificial Analysis measured Terra at 48.1 on its Intelligence Index at maximum effort, below GPT-6.1 Sol, GPT-6 Astra and several other models in its comparison.
Which Terra effort setting should users choose?
The source reports that low and medium effort offer the strongest cost efficiency in Artificial Analysis’s evaluation. Users should test settings on their own work: more effort costs more, and harder tasks may require a larger model.
Are Terra’s speed and safety claims independently verified?
Not in the supplied material. OpenAI says Terra is more than three times faster than Sol and Astra and meets a Critical cybersecurity threshold under its framework. The cited independent evaluation had not published Terra speed measurements, and the safety classification is OpenAI’s own assessment.
Source: ThorstenMeyerAI.com
