🔍 Read the full analysis: The Appeal Of GPT-6.1 Terra: Useful AI Without The Highest Price on ThorstenMeyerAI.com
Get smart everyday buys 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 everyday work and computer-based tasks. Artificial Analysis measured task costs from $0.05 to $0.40 across effort settings, with low and medium effort on its cost-efficiency frontier; Terra still scores below several higher-performing models, and some cheaper alternatives outscore its highest settings.
OpenAI released GPT-6.1 Terra on October 26, describing it as a lower-cost model for work and computer-based tasks. Its API price is $1 per million input tokens and $5 per million output tokens, half the listed rates for GPT-5.6 Terra, while independent testing found that its cost per completed task depends sharply on the selected reasoning effort.
Terra is positioned between OpenAI’s smaller Luna model and its larger Sol and Astra models. OpenAI says it is available to eligible 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. The listed price for cached input is $0.05 per million tokens. OpenAI says the same ChatGPT allowance can cover substantially more Terra tasks than work with Sol or Astra.
Artificial Analysis published an independent evaluation on launch day. At maximum effort, Terra scored 48.1 on the firm’s Intelligence Index and cost an average of $0.40 per task across that index’s evaluations. By comparison, GPT-6.1 Sol scored 51.8 at $0.72 per task, and Claude Opus 5.5 scored 57.6 at $5.98. These are averages tied to Artificial Analysis’s benchmark tasks, not a guaranteed cost for a particular user’s workload.
The evaluation’s six effort settings show a substantial trade-off: scores range from 33.5 to 48.1, while average task costs range from $0.055 at low effort to $0.40 at maximum effort. Artificial Analysis places low and medium effort on its cost-efficiency frontier, meaning it found no tested model that was both cheaper and better at those points. The supplied comparison also says Terra’s non-reasoning mode costs nearly twice as much per task as low effort while scoring lower.
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 practical distinction for organizations and individuals buying AI: token price is not the same as completed-task cost. A model with a lower price per token may use more tokens to finish a job, while a model with a higher rate may complete it with fewer. Artificial Analysis’s figures suggest that Terra at low effort can be cheaper per task than Luna, despite Luna’s lower listed token price. That result depends on the benchmark and should not be assumed for every use case.
For companies handling repeated work such as invoice processing, the relevant budget may be the cost of reliably finishing each task, alongside the time and review required. OpenAI identifies reading invoices and entering their details into accounting software as an example of the high-volume work Terra is meant to handle. Its claim that Terra completes tasks more than three times faster than Astra and Sol is based on OpenAI’s own measurements; Artificial Analysis had not published speed data for Terra at the time of the source report.
Terra’s position is not simply “cheap and best.” Artificial Analysis reported higher scores for Claude Opus 5.5, Kimi K3.1 Pro and other models. It also listed Xiaomi’s open-weight MiMo-V2.6-Pro at 46.3 points and $0.13 per task, a lower cost and higher score than Terra’s high and xhigh effort settings, according to the source’s comparison. Buyers therefore need to weigh task quality, model access, speed and review needs rather than rely on one price or score.
Products Worth Considering
Logitech MX Mechanical Full-Size Wireless Mechanical Keyboard, Tactile Quiet Switches, Backlit Keys, Bluetooth, USB-C, Multi-Device, Graphit
Dell UltraSharp 27 Thunderbolt Hub Monitor (U2726DER)
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
As an affiliate, we earn on qualifying purchases.
Terra Between Luna and Sol
OpenAI describes Terra as a model for “the communication and computer-based tasks that fill the workday.” Its launch materials present the model as a middle option in its lineup: more capable than Luna on some measures, but below Sol and Astra for the most demanding work. OpenAI says Sol and Astra remain the better choice for the most complex coding tasks.
The published comparisons cover several different tests, and their results should not be treated as directly interchangeable. On Artificial Analysis’s Intelligence Index, which combines ten evaluations, GPT-6.1 Terra scored 48.1 at maximum effort. In the firm’s separate Coding Agent Index, Terra scored 48, compared with 54 for Sol and 31 for Luna. OpenAI separately reported Terra solving 50.3% of tasks on Terminal-Bench 4.0 at maximum effort, versus 21.5% for GPT-5.6 Terra’s published result.
OpenAI also reported improved error rates on its own set of error-inducing prompts: at low effort, the rate fell from 53.6% for GPT-5.6 Terra to 16.7% for GPT-6.1 Terra; at maximum effort, it fell from 22.3% to 4.3%. Those results come from OpenAI’s internal evaluation, not an independent test. Artificial Analysis separately found that Terra guessed on 38% of questions it could not answer in its hallucination test, down from 88% for GPT-5.6 Terra. The two measures assess different tests and should remain distinct.
Limits of the Published Comparisons
The launch-day comparisons do not establish how Terra will perform on every customer’s workload. Benchmark scores and average costs depend on test design, task mix, reasoning effort and token use. The supplied material does not provide a complete breakdown of all six effort settings, including every setting’s cost and score, so the reported range should not be used to infer missing values.
Several performance claims come from OpenAI’s own tests, including its speed comparison, computer-use results, coding figures and error-rate measurements. The source says Artificial Analysis had not published Terra speed measurements when the report was prepared. It is also unclear from the material how Terra’s observed costs and quality will vary across real deployments, or what review burden customers may need for tasks with factual or operational consequences.
OpenAI says Terra meets the “Critical” threshold in cybersecurity under its Preparedness Framework and retains safeguards developed for GPT-6 Astra. That classification is OpenAI’s own; the source does not provide an independent evaluation of the claim.
What Buyers Should Measure Next
The next practical test is how Terra performs on customers’ own tasks. Organizations considering it can compare cost per successfully completed task, error rates, speed and the amount of human checking needed across effort settings. Starting with low or medium effort is consistent with Artificial Analysis’s cost-efficiency finding, but the appropriate setting will depend on the task and the consequences of a mistake.
Further independent testing could clarify Terra’s speed, real-world reliability and performance across specialized workloads. Until those results are available, OpenAI’s launch figures and Artificial Analysis’s benchmark provide different kinds of evidence: company-reported results for selected tests and an independent index covering its own evaluation set. The decision for users is whether Terra’s lower price and performance are a fit for their specific work, not whether it replaces higher-capability models across the board.
Key Questions
How much does GPT-6.1 Terra cost?
OpenAI lists API prices 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 benchmark task at maximum effort; task cost varies with effort and workload.
Where can users access Terra?
According to the source, Terra is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and through the API as gpt-6.1-terra.
Is Terra OpenAI’s highest-performing model?
No. The reported independent benchmark scores place Terra below Sol and several competing models. OpenAI also says Sol and Astra remain the better choice for its most complex coding tasks.
Which Terra effort setting is cheapest?
In Artificial Analysis’s reported measurements, low effort cost $0.055 per task, compared with $0.107 for non-reasoning mode and $0.40 at maximum effort. Those are averages for the firm’s evaluation tasks, not fixed prices for every request.
What remains unverified?
The supplied material does not include independent speed measurements or results from every real-world workload. OpenAI’s internal test results and safety classification are company-reported, while Artificial Analysis’s figures apply to its own benchmark tasks.
Source: ThorstenMeyerAI.com
