🔍 Read the full analysis: When Does GPT-6.1 Terra Make Sense For AI Tasks? on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- 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 work. Artificial Analysis found its low- and medium-effort settings on the cost-efficiency frontier in its tested models, while higher effort costs more without matching the strongest models’ scores. Results vary by task, and the source’s six-setting measurements are incomplete.
OpenAI released GPT-6.1 Terra on October 26, 2026, offering a mid-tier model for routine AI work at lower API token prices than its predecessor. In an evaluation published the same day, Artificial Analysis found that Terra’s low- and medium-effort settings sit on its cost-efficiency frontier: among tested models, no alternative was both cheaper per task and higher-scoring. That makes Terra a candidate for repeatable, high-volume tasks, not a universal replacement for stronger models.
OpenAI lists API pricing at $1 per million input tokens and $5 per million output tokens, down from GPT-5.6 Terra’s $2 and $12. Cached input costs $0.05 per million tokens. The model is available in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, as well as through the API under the name gpt-6.1-terra. OpenAI describes it as a fast, efficient model for communication and computer-based work, citing invoice entry as an example.
Artificial Analysis measured Terra’s Intelligence Index score at 48.1 at maximum effort, with an average cost of $0.40 per task across its evaluation tasks. 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 are averages from the benchmark’s task mix, not a price guarantee for every real-world job. The index combines ten evaluations, including agentic office work, terminal tasks, science and hallucination.
OpenAI’s own figures address different tests and should not be conflated with Artificial Analysis’s results. On Agents’ Last Exam, OpenAI reports Terra at medium effort scored 57.6% for $0.29 per task; at maximum effort it reports 60.0% for $1.01. OpenAI also reports 50.3% on Terminal-Bench 4.0 at maximum effort. The company 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.
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
Where Terra Cuts Task Costs
The pricing comparison depends on what is being measured. Cost per token tells buyers the rate for generated or processed text; cost per task also reflects how many tokens a model uses to finish a job. Artificial Analysis found that GPT-6 Luna, despite its lower token price, cost more per task than Terra at low effort in the reported comparison, because Terra used fewer tokens. For organizations processing many similar requests, that difference can affect budgets and how many tasks a usage allowance covers.
Terra’s measured trade-off is between capability and expense. The source reports six effort settings spanning scores of 33 to 48 and task costs of $0.05 to $0.40. Maximum effort cost 7.4 times as much as low effort for 12 more index points. On those measurements, low and medium effort were on the cost-efficiency frontier; higher settings were not. The practical implication is to test the cheapest setting that reliably meets a task’s accuracy requirements, rather than automatically selecting maximum effort.
That does not make Terra the best choice for every job. Opus 5.5, Kimi K3.1 Pro and Gemini 4 Argon scored higher in the cited Artificial Analysis comparison. For complex coding, OpenAI itself says Sol and Astra remain better choices. A modest per-task saving may not be worthwhile if the lower-cost model needs human correction or fails a task that a stronger model can complete.
Popular right now
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black
Vitamix Propel Series 510 Professional-Grade Blender, 3 Programs, Black
Vitamix 5200 Professional-Grade Blender, 64 oz. Container, Black
As an affiliate, we earn on qualifying purchases.
How Terra Fits OpenAI’s Lineup
OpenAI positions Terra between GPT-6 Luna, its smaller model, and the larger Sol and Astra models. Its listed API prices are 50% lower than GPT-5.6 Terra’s, and the company says the same ChatGPT Work allowance can cover substantially more tasks than with Sol or Astra. Those are OpenAI’s product and pricing claims; the amount a particular user completes will depend on task length and usage.
Artificial Analysis’s launch-day assessment supplies a separate comparison. At maximum effort, it scored Terra 48.1, four points below Sol and roughly ten points above Luna. The source also reports a lower rate of guessed answers on Artificial Analysis’s hallucination test: Terra guessed on 38% of questions it could not answer, compared with 88% for GPT-5.6 Terra. That is a result on a specific benchmark, not proof that Terra will avoid unsupported answers in every setting.
Some favorable results come from OpenAI’s own testing. The company reports lower error rates on its internal error-inducing prompts and stronger performance than GPT-5.6 Terra on the cited coding test. These measures provide useful evidence about the tasks tested, but they are vendor-reported; independent benchmark results and performance on customers’ own workloads may differ.
Limits of the Available Benchmarks
The results do not establish how Terra will perform on every company’s data, workflows or quality standards. Benchmark scores reflect selected tests, and cost-per-task averages depend on the evaluation mix and the number of tokens used. The source material gives only partial details for the six effort settings: it identifies the range and several endpoint figures, but does not provide a complete setting-by-setting table. It is not possible from those details alone to verify every intermediate score, cost or hallucination rate.
Independent speed measurements were not available in the cited assessment, so OpenAI’s claim that Terra is more than three times faster than Sol and Astra remains a vendor claim in this account. The material also does not specify how Terra’s per-task costs will vary across customer workloads, or whether every ChatGPT plan’s allowance treats the model identically. OpenAI’s cybersecurity classification is likewise its own assessment under its Preparedness Framework, not an independent finding described here.
Test Terra Against Real Workloads
The next useful comparison is likely to come from independent testing and customer deployments that publish results on specific workloads, including completion rates, correction needs, latency and total cost. Organizations considering Terra can begin with low or medium effort on routine tasks, then compare the output against an existing model and measure the full cost of accepted work, including review. More demanding jobs can be routed to Sol, Astra or another stronger model when Terra does not meet the required standard.
For individual users, Terra’s lower listed token price may stretch usage, but the benefit depends on task complexity and the model’s output. OpenAI’s release and the launch-day benchmark establish a reason to test Terra as a cost-conscious option; they do not settle which model is best for a particular user. Independent speed data, fuller effort-setting comparisons and results from real deployments remain useful next evidence.
Key Questions
When was GPT-6.1 Terra released?
OpenAI released GPT-6.1 Terra on October 26, 2026. Artificial Analysis published an evaluation on launch day.
What 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 tokens. Actual task cost depends on how many tokens a job uses.
When does Terra make sense for an AI task?
The cited benchmark suggests testing Terra’s low or medium effort for routine, repeatable work where cost matters and the model meets the required quality bar. Teams should measure cost and correction needs on their own tasks.
Is Terra better than GPT-6.1 Sol or Claude Opus 5.5?
Not by the reported maximum-effort Intelligence Index scores: Terra scored 48.1, versus 51.8 for Sol and 57.6 for Opus 5.5. Terra’s average benchmark task cost was lower, so the choice depends on the balance of price and capability needed.
Does Terra replace OpenAI’s larger models for coding?
No. OpenAI reports a 50.3% score for Terra on Terminal-Bench 4.0 at maximum effort, but says Sol and Astra remain better choices for its most complex coding tasks.
Source: ThorstenMeyerAI.com
