🔍 Read the full analysis: GPT-6.1 Terra Could Put Cost-Effective AI To Work 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 workplace and computer-based tasks. Artificial Analysis measured its cost at $0.05 to $0.40 per task across effort settings, but Terra’s top-setting score remained below several leading models, and some cheaper alternatives scored better.
OpenAI released GPT-6.1 Terra on October 26 as a lower-cost model for workplace and computer-based tasks, pricing its API at $1 per million input tokens and $5 per million output tokens. Independent benchmarking firm Artificial Analysis found that the cost of completing its evaluation tasks ranged from $0.05 to $0.40 depending on Terra’s effort setting, while its top-setting score remained below several higher-performing models.
OpenAI describes Terra as a “fast, efficient model” for routine communication and computer work. The company’s examples include reading incoming invoices and entering their details into accounting software. Terra is available through the API as gpt-6.1-terra and to Plus, Pro, Enterprise and Edu subscribers in ChatGPT Work and Codex. OpenAI says the same allowance can cover substantially more tasks than when using its Sol or Astra models.
The listed API rate is 50% lower than GPT-5.6 Terra’s $2-per-million input and $12-per-million output price. Cached input tokens cost $0.05 per million, according to the launch material. OpenAI also reported that Terra completed tasks more than three times faster than Sol and Astra, but Artificial Analysis had not published independent speed measurements for Terra when the source data was checked.
Artificial Analysis gave Terra an Intelligence Index score of 48.1 at maximum effort, with an average evaluation-task cost of $0.40. The firm’s index combines ten evaluations, including office work, terminal tasks, science and hallucination tests. For 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 performance and average task costs on that benchmark, not the price of every real-world job.
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
Task Costs Change the Model Choice
Terra’s launch highlights a practical distinction for companies buying AI services: token price is not the same as task cost. A model that charges less per token can still use more tokens to finish a job. Artificial Analysis found GPT-6 Luna’s maximum setting was much cheaper per token, but Terra at low effort cost less per task in its evaluation because it generated fewer tokens.
The benchmark’s effort curve also shows why choosing a setting matters. Artificial Analysis measured Terra from 33.5 points at non-reasoning to 48.1 at maximum effort, with average task costs rising from $0.107 to $0.40. Low effort scored 36.2 at $0.055 per task; medium effort was also on the firm’s cost-efficiency frontier, meaning no tested model in its comparison was both cheaper and better at those points. The higher settings did not meet that definition.
For organizations processing large volumes of routine work, a lower cost per completed task can make automation more affordable, provided the model performs reliably on their own data. Terra is not a universal substitute for larger models: OpenAI says Sol and Astra remain better choices for its most complex coding tasks. Buyers may need to route simple work to Terra and reserve more capable models for harder cases, then compare costs and error rates in production.
Products Worth Considering
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
DJI Osmo Mobile 7P with Mic Mini TX (Infinity Black) Gimbal Stabilizer
Bose QuietComfort Headphones (2nd Gen), Wireless Noise-Cancelling Over-Ear Headphones, Eucalyptus Green
As an affiliate, we earn on qualifying purchases.
How Terra Compares on Benchmarks
Artificial Analysis’s maximum-effort ranking places Terra below several models with higher index scores, including Claude Opus 5.5, Kimi K3.1 Pro, GPT-6 Astra, Gemini 4 Argon and GPT-6.1 Sol. It also lists lower-cost competitors: Xiaomi’s open-weight MiMo-V2.6-Pro scored 46.3 at an average $0.13 per task, while DeepSeek V4.1 Flash scored 45.1 at $0.26. Results depend on the evaluation set and settings, so the index is one comparison rather than a guarantee of performance on a company’s specific workload.
OpenAI’s launch materials report separate results on its selected tests. On Agents’ Last Exam, Terra at medium effort scored 57.6% for $0.29 per task, compared with 42.6% and $0.84 for GPT-5.6 Terra at medium effort; OpenAI said that result matched GPT-6 Astra at medium effort, which cost $4.82 per task in its chart. On Terminal-Bench 4.0, OpenAI reported a 50.3% solve rate for Terra at maximum effort, versus 21.5% for GPT-5.6 Terra’s published result. These are company-reported figures, distinct from Artificial Analysis’s evaluation.
OpenAI also reported lower error rates on an internal set of prompts designed to induce factual mistakes: 16.7% at low effort and 4.3% at maximum, down from 53.6% and 22.3% for GPT-5.6 Terra at the corresponding settings. In Artificial Analysis’s hallucination test, Terra guessed on 38% of questions it could not answer, compared with 88% for GPT-5.6 Terra. The tests use different methods and should not be treated as a single direct measure of accuracy.
Limits of the Cost and Score Data
The benchmark results do not establish how Terra will perform across every business workflow. Artificial Analysis’s task-cost figure is an average over its evaluation set; actual expenses can vary with prompt length, output size, cached inputs, reasoning effort and the work being assigned. The available source material does not provide independent speed measurements or a full, independently verified comparison of Terra’s performance in live deployments.
OpenAI’s accuracy, speed, computer-use and cybersecurity statements are company-reported claims, while Artificial Analysis supplies an independent benchmark with its own methods. The materials also do not specify how the reported scores translate into error rates for particular customers, or whether the same results hold across languages and specialized tasks. OpenAI says Terra meets its Preparedness Framework’s “Critical” threshold in cybersecurity and retains safeguards developed for GPT-6 Astra; that classification is OpenAI’s own, not an independent assessment presented here.
The supplied effort-setting data shows a broad trade-off between cost and score, but does not provide a complete breakdown for every setting across all benchmark measures. Users should not infer that a particular setting will always be cheapest or most accurate for their own task without testing it.
Testing Terra in Live Workflows
Terra is available now through the API, ChatGPT Work and Codex for the listed subscriber plans. The next practical test is whether its lower benchmark task costs carry over to real workflows, including invoice processing, office administration and computer-use tasks. Companies evaluating it can compare the cost of completed work—not just token rates—alongside accuracy, correction time and the need for human review.
Further independent measurements, especially on speed and task-specific performance, could clarify where Terra sits against both OpenAI’s larger models and lower-priced competitors. OpenAI’s own guidance points to a mixed approach: use Terra for suitable routine work and retain Sol or Astra for more complex coding. No adoption figures, customer results or additional benchmark release schedule were included in the launch information.
Key Questions
What is GPT-6.1 Terra?
GPT-6.1 Terra is an OpenAI model marketed for workplace communication and computer-based tasks. It is available through the API, ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu subscribers.
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. Artificial Analysis measured average costs of $0.05 to $0.40 per evaluation task across Terra’s effort settings; that is a benchmark average, not a fixed charge for every task.
Is Terra OpenAI’s most capable model?
No. Artificial Analysis scored Terra 48.1 at maximum effort, below GPT-6.1 Sol at 51.8 and several other tested models. OpenAI says Sol and Astra remain the better options for its most complex coding tasks.
Why compare cost per task instead of token price?
A model’s total cost depends on how many input and output tokens it uses to finish the work. Artificial Analysis found that Terra at low effort could cost less per task than a model with a lower token price, because Terra used fewer tokens in its evaluation.
Source: ThorstenMeyerAI.com
