🔍 Read the full analysis: GPT-6.1 Terra May Trade Peak Smarts For Lower AI Costs 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, positioning it as a lower-cost model for routine work. Artificial Analysis measured a $0.40 average task cost and a 48.1 Intelligence Index score at maximum effort; higher effort costs more, while the model trails several larger rivals.
OpenAI released GPT-6.1 Terra on October 26, describing it as a fast, efficient model for routine communication and computer-based work. Its API rates are $1 per million input tokens and $5 per million output tokens, half the listed rates for its predecessor, GPT-5.6 Terra. An independent evaluation found Terra can complete benchmark tasks at lower average cost than some larger models, while scoring below leading systems on the same index.
Terra is available through the API under the model name gpt-6.1-terra and to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, according to the launch report. OpenAI says its paid ChatGPT plans can complete substantially more tasks with the same allowance than when using Sol or Astra. The company presents invoice processing—reading incoming invoices and entering details into accounting software—as an example of the high-volume work Terra is intended to handle.
Artificial Analysis gave Terra an Intelligence Index score of 48.1 at maximum effort, at an average cost of $0.40 per task across its evaluation set. The firm reported scores of 51.8 and costs of $0.72 for GPT-6.1 Sol, and 57.6 and $5.98 for Claude Opus 5.5. Its figures are average task costs within that benchmark, not a guaranteed price for every user’s workload. At maximum effort, Terra’s score remained below those models despite its lower measured cost.
Terra has six effort settings, with Artificial Analysis measurements spanning scores from 33.5 to 48.1 and average costs from $0.055 to $0.40 per task. The report says low and medium effort fall on its cost-efficiency frontier; higher settings cost more without matching the same cost-to-score trade-off. The supplied launch report does not include the complete figures for every setting, so comparisons beyond the listed measurements should be treated cautiously.
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 Shape Terra’s Appeal
The release puts attention on cost per completed task, rather than token rates alone. A model’s real expense depends on how many input and output tokens it uses to finish the work, along with its reasoning behavior. The source reports that GPT-6 Luna costs less per token, but Terra at low effort costs less per task in Artificial Analysis’s measurements because it generated fewer tokens on the tested work. That is a result from a specific benchmark, not a universal outcome for every prompt or business process.
For organizations handling large volumes of repetitive work, a lower average task cost could make AI use more affordable, especially when tasks do not need a top-scoring model. But benchmark performance still matters: the source says OpenAI considers Sol and Astra better choices for its most complex coding tasks. Companies would need to test Terra against their own tasks, error tolerance and usage patterns before relying on the benchmark comparison.
The measured trade-off also gives users a reason to select effort settings deliberately. Low or medium effort may suit routine requests, while difficult work may benefit from additional reasoning or a more capable model. More reasoning is not automatically better value: the maximum setting costs more than low effort, and the benchmark’s additional score must be weighed against that expense.
Products Worth Considering
LG UltraGear 45GX900A-B 45-Inch Curved OLED Gaming Monitor
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
LG UltraGear 32GX870A-B 32-inch 4K OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
How Terra Compares With Rivals
OpenAI’s launch claims and the independent evaluation measure different things. OpenAI cited internal tests and selected benchmarks, while Artificial Analysis combined ten evaluations in its Intelligence Index, covering areas such as agentic office work, terminal tasks, science and hallucination. Its cost figure reflects average spending across those evaluation tasks and includes cached input and reasoning tokens, according to the source report.
OpenAI said Terra’s API prices are 50% lower than GPT-5.6 Terra’s and match Sol’s listed per-token rates at half the price. Artificial Analysis also found Terra’s maximum-effort task cost below Sol’s in its own evaluation. These are not interchangeable comparisons: one is a token price, while the other is an average cost across benchmark tasks. The latter can vary with task design and how much a model writes.
The release report also cites OpenAI’s own results for computer use, factual errors and coding. OpenAI said Terra scored 57.6% on Agents’ Last Exam at medium effort, compared with 42.6% for GPT-5.6 Terra, and reported lower error rates on its internal error-inducing prompts. On Terminal-Bench 4.0, OpenAI reported a 50.3% result at maximum effort for Terra, compared with 21.5% for GPT-5.6 Terra. These are company-reported figures and should not be confused with independent findings.
Benchmark Limits and Open Questions
The figures do not establish how Terra will perform or what it will cost on any particular company’s workload. Benchmark averages are not user-specific estimates, and the source material does not provide complete measurements for every effort setting or a full account of how results may change across different task mixes. Artificial Analysis had not published speed measurements for Terra when the report was written, despite OpenAI’s claim that Terra completes tasks more than three times faster than Astra and Sol.
Some reported results are claims from OpenAI’s internal tests or launch materials, rather than independently verified comparisons. The source says OpenAI classified Terra as meeting the Critical threshold in cybersecurity under its Preparedness Framework, but provides no independent assessment of that classification. Details about real-world error rates, deployment safeguards and performance at different usage levels remain limited in the material provided.
The available report is also incomplete: it says the non-reasoning setting costs almost twice as much per task as low effort while scoring lower, but does not provide the full six-setting table. The exact cost and performance of each intermediate setting cannot be established from the supplied figures alone.
Testing Terra on Real Workloads
Terra is available now through the listed ChatGPT plans, Codex and the API, according to the launch report. The next useful evidence will come from independent speed measurements, fuller results across all six effort settings, and comparisons on real-world tasks rather than benchmark averages alone. Companies considering deployment will need to measure both the cost and accuracy of completed tasks in their own workflows.
OpenAI’s launch materials suggest starting with lower-effort settings for routine tasks and using larger models for complex coding. That is the company’s positioning, not a guarantee that one routing strategy will fit every user. Further comparisons from Artificial Analysis and other evaluators may clarify where Terra’s lower measured costs hold up and where higher-scoring alternatives justify their additional expense.
Key Questions
What is GPT-6.1 Terra?
GPT-6.1 Terra is a model OpenAI released on October 26, 2026, for communication and computer-based tasks. It is positioned between the smaller Luna model and the larger Sol and Astra models.
How much does GPT-6.1 Terra cost?
The source lists API rates of $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 task at maximum effort across its benchmark, but that is not a price guarantee for other workloads.
Does Terra outperform OpenAI’s larger models?
Not on the Artificial Analysis Intelligence Index figures cited in the report. Terra scored 48.1 at maximum effort, below GPT-6.1 Sol at 51.8 and GPT-6 Astra at 52.7. OpenAI says Sol and Astra remain better choices for its most complex coding work.
Why compare cost per task instead of token price?
Token rates do not show how many tokens a model uses to finish a task. Cost per task includes the tokens used in a particular evaluation, so a model with a higher rate per token can still cost less on a task if it uses fewer tokens. The result depends on the work being measured.
Source: ThorstenMeyerAI.com
