🔍 Read the full analysis: GPT-6.1 Terra And The Case For A More Affordable OpenAI Model on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
OpenAI released GPT-6.1 Terra on October 26, 2026, positioning it as a lower-cost model for routine work. Artificial Analysis found that Terra’s low- and medium-effort settings offer favorable cost per task, but higher-scoring models remain available, and the source data does not establish that Terra is best for every workload.
OpenAI released GPT-6.1 Terra on October 26, 2026, offering a model priced at $1 per million input tokens and $5 per million output tokens for routine work such as office and computer-use tasks. Artificial Analysis’s launch-day testing found that Terra’s low- and medium-effort settings sit on its cost-efficiency frontier, while the model’s maximum-effort score remains below several competing systems.
OpenAI says Terra is designed for “communication and computer-based tasks” that make up a workday. It is available in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and through the API as gpt-6.1-terra. OpenAI lists cached input at $0.05 per million tokens and says the same ChatGPT allowance can cover substantially more tasks than with Sol or Astra.
The listed API price is half the per-token price of GPT-5.6 Terra, which cost $2 per million input tokens and $12 per million output tokens, and below GPT-6.1 Sol’s $2 and $10 rates. But token price alone does not show the cost of completing a job. Artificial Analysis measured Terra’s average cost across its evaluation tasks at $0.40 at maximum effort, compared with $0.72 for Sol and $5.98 for Claude Opus 5.5. Terra scored 48.1 on the firm’s Intelligence Index, versus 51.8 for Sol and 57.6 for Opus.
Artificial Analysis tested six effort settings, with scores ranging from 33.5 in non-reasoning mode to 48.1 at maximum effort and costs from about $0.05 to $0.40 per task. Its assessment places the low and medium settings on the cost-efficiency frontier; settings above those points cost more without a better score-to-cost position. The provided figures also show non-reasoning mode costs $0.107 per task, nearly twice the low setting’s $0.055, while scoring lower. The evaluation uses an average across benchmark tasks, not a guarantee of what any particular customer will pay.
OpenAI’s own launch-page data reports improvements on selected tests. On Agents’ Last Exam, Terra at medium effort scored 57.6%, compared with 42.6% for GPT-5.6 Terra, at a reported $0.29 rather than $0.84 per task. On Terminal-Bench 4.0, Terra at maximum effort solved 50.3% of tasks, against 21.5% for GPT-5.6 Terra’s published result. These are company-reported benchmark results; they are separate from Artificial Analysis’s evaluation.
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 Choice
Terra’s launch highlights a practical purchasing question for organizations using AI at scale: the cost to finish a task can matter more than the posted price per token. A cheaper token rate does not automatically mean a cheaper completed job, because models may differ in how many tokens they use. Artificial Analysis’s figures show Luna, for example, has a lower listed token price, yet Terra at low effort costs less per evaluation task in the reported comparison.
For businesses processing invoices, handling computer-based workflows or running coding agents, a lower per-task bill could make automation more affordable if the model performs adequately on the specific job. The trade-off is capability: Terra’s maximum-effort score is below Opus 5.5, Sol and several other tested models, and OpenAI says Sol and Astra remain better choices for its most complex coding tasks. The useful takeaway is not that Terra replaces larger models, but that model choice and reasoning effort can be matched to task difficulty.
Artificial Analysis also found that a lower effort setting can provide a stronger cost-to-score balance than either no reasoning or maximum effort. That makes workload testing important: a company may need to compare accuracy, completion rates and total spending on its own tasks before routing work to Terra. The benchmark’s average costs are a reference point, not an enterprise-specific forecast.
Popular right now
TCL 85 Designer Series QD-Mini LED TV (Model 85A400U) with Google Smart TV
Vitamix 5200 Professional-Grade Blender, 64 oz. Container, Black
Shark Navigator Robot Vacuum with Self-Empty Base, Dark Grey (RV2120AE)
As an affiliate, we earn on qualifying purchases.
Where Terra Fits in OpenAI’s Lineup
OpenAI presents Terra between its smaller Luna model and its larger Sol and Astra models. The launch material describes Terra as a more efficient option for common work, including reading incoming invoices and entering details into accounting software. OpenAI says it is more than three times faster than Sol and Astra, but Artificial Analysis had not published speed measurements for Terra at the time covered by the source.
Artificial Analysis’s Intelligence Index combines ten evaluations covering areas such as agentic office work, terminal tasks, science and hallucination. Its launch-day figures put Terra at 48.1 points and $0.40 per task at maximum effort. The index is one independent benchmark, not a full measure of how the model performs in every business setting. The comparison also includes Xiaomi’s open-weight MiMo-V2.6-Pro, which scored 46.3 at a reported $0.13 per task, illustrating that lower-cost alternatives exist.
OpenAI reports lower error rates on its internal error-inducing prompts: 16.7% at low effort versus 53.6% for GPT-5.6 Terra, and 4.3% at maximum versus 22.3%. 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. These figures come from different tests and should not be treated as directly interchangeable. OpenAI also says Terra meets its “Critical” cybersecurity threshold under its Preparedness Framework; that is the company’s own classification.
Questions Before Wider Deployment
It is not yet clear how Terra will perform across customers’ real-world tasks, which may differ from the evaluations used by Artificial Analysis and OpenAI. The source material does not provide independent speed measurements, detailed customer results or a breakdown of how task costs change across different production workloads. OpenAI’s claims about faster completion and its internal benchmark improvements have not been independently verified in the information provided.
Benchmark costs also depend on each test’s prompts, token use and evaluation mix. Artificial Analysis’s average is informative for comparisons under its methodology, but it does not establish a fixed cost for a business task. The supplied source excerpt gives only some of the six effort-setting figures, so the exact scores and costs for every intermediate setting cannot be reported here. It also remains unclear how often customers will need to move from Terra to Sol or Astra when tasks are difficult or errors carry higher costs.
Testing Terra on Real Workloads
Organizations considering Terra can compare it with their current models on representative jobs, tracking completion quality, error rates, latency and total cost per finished task. For routine work, the reported low- and medium-effort results provide a starting point, but settings should be evaluated against the required accuracy rather than chosen on price alone.
Further independent testing, including speed measurements and results from deployed systems, would clarify how Terra performs beyond benchmark suites. OpenAI’s product availability means eligible ChatGPT users and API customers can assess the model now, while the durability of the reported savings and the best way to route tasks between Terra, Sol and Astra remain open questions.
Key Questions
What is GPT-6.1 Terra?
GPT-6.1 Terra is an OpenAI model released on October 26, 2026, for routine communication and computer-based tasks. OpenAI offers it in ChatGPT Work and Codex for eligible plans and through its API.
How much does 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 tokens. Artificial Analysis measured an average cost of $0.40 per evaluation task at maximum effort; its low-effort figure was $0.055.
Is Terra better than OpenAI’s larger models?
Not across the board. Artificial Analysis scored maximum-effort Terra at 48.1, below GPT-6.1 Sol at 51.8 and several other models. OpenAI says Sol and Astra remain better choices for its most complex coding tasks.
Why compare cost per task instead of token price?
Models can use different amounts of text and reasoning tokens to complete a job. A lower token price does not necessarily produce a lower total bill, so cost per completed task can be more useful when comparing systems for a particular workload.
Source: ThorstenMeyerAI.com
