🔍 Read the full analysis: Is Paying For OpenAI’s Smartest Model Necessary? Consider GPT-6.1 Terra 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 routine work rather than its most capable option. Artificial Analysis measured a $0.40 average task cost and a 48.1 Intelligence Index score at maximum effort; its results suggest low and medium effort offer stronger cost-efficiency, while harder tasks may benefit from larger models.
OpenAI released GPT-6.1 Terra on October 26 as a lower-cost model for everyday computer-based work, rather than as its top-performing system. Independent benchmark firm Artificial Analysis measured Terra at a 48.1 Intelligence Index score and an average cost of $0.40 per task at maximum effort, below the scores of several larger models but at a fraction of their task costs.
OpenAI lists Terra’s API price at $1 per million input tokens and $5 per million output tokens, compared with $2 and $12 for GPT-5.6 Terra. Cached input is listed at $0.05 per million tokens. The model is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, as well as through the API under the name gpt-6.1-terra. OpenAI says the same allowance can cover substantially more Terra tasks than work using Sol or Astra.
Artificial Analysis’s evaluation combines ten tests, including agentic office work, terminal tasks, science and hallucination, and accounts for task costs including reasoning tokens. At maximum effort, Terra scored 48.1 for $0.40 per task. GPT-6.1 Sol scored 51.8 for $0.72, while Claude Opus 5.5 scored 57.6 for $5.98. The figures are averages across the benchmark’s evaluation tasks, not a guaranteed cost for any individual user’s work.
OpenAI also reported results from its own tests. On Agents’ Last Exam, it said Terra at medium effort scored 57.6% and cost $0.29 per task, compared with 42.6% and $0.84 for GPT-5.6 Terra at medium effort. 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-provided measurements and should be distinguished from the independent Artificial Analysis 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
The Cost of Finishing a Task
The release puts a practical purchasing question ahead of a simple model ranking: what does a completed task cost? Token prices alone can mislead because models may use different amounts of text and reasoning to reach an answer. Artificial Analysis found GPT-6 Luna, a smaller model, costs ten times less per token than Terra, yet Terra at low effort cost less per task in the benchmark because it generated fewer tokens. That comparison is specific to the tested tasks; costs will vary with prompts, workloads and settings.
For businesses handling many routine requests, lower per-task costs could make automation more affordable, particularly for work such as processing invoices. But Terra’s lower price does not make it a universal substitute for larger models. Artificial Analysis placed it below several higher-scoring systems, and OpenAI says Sol and Astra remain better choices for the most complex coding work. The relevant decision is a trade-off between cost, reliability and task difficulty, not simply whether a model is cheaper.
Artificial Analysis reported that Terra guessed on 38% of questions it could not answer in its hallucination test, compared with 88% for GPT-5.6 Terra. That result suggests improvement on that particular measure, not an absence of fabricated answers. Users and organizations still need checks for consequential outputs, especially where errors could affect customers, finances or records.
Products Worth Considering
Apple iPad Air (2025) 13‑inch Wi‑Fi 256 GB – Space Gray (M3 chip)
FLEXISPOT 36-Inch Electric Standing Desk Converter, Black
Apple iPad Pro 11-inch (M5), 256GB, Wi-Fi 7, Space Black
As an affiliate, we earn on qualifying purchases.
Where Terra Sits in OpenAI’s Lineup
OpenAI describes Terra as a “fast, efficient model” for communication and computer-based tasks that fill the workday. The company’s example is reading incoming invoices and entering them into accounting software. Terra sits between the smaller Luna model and the larger Sol and Astra models in OpenAI’s lineup, with six selectable effort settings that range from non-reasoning to maximum effort.
Artificial Analysis measured Terra’s Intelligence Index score from 33.5 at non-reasoning to 48.1 at maximum effort. Its reported task cost ranged from $0.107 at non-reasoning to $0.40 at maximum. Low effort scored 36.2 at $0.055 per task, while medium effort scored higher at a higher cost. The source’s summary says low and medium effort sit on its cost-efficiency frontier: among tested options at those points, no model was both cheaper and better. The settings above them did not retain that position.
The steepest setting is not automatically the best value. Artificial Analysis’s summary says maximum effort costs 7.4 times as much as low effort for 12 additional index points. The same evaluation found the non-reasoning mode was not the least expensive setting per task: it cost almost twice as much as low effort while scoring lower. These figures show why users should compare complete-task results at the effort level they expect to use, rather than assuming less reasoning always means lower cost.
What the Benchmarks Cannot Establish
The published results do not establish how Terra will perform across every company’s software, documents or user prompts. Artificial Analysis’s task cost is an average across its evaluation set; actual spending depends on the work submitted, the effort setting and token use. The source material also says Artificial Analysis had not published speed measurements for Terra when its evaluation was checked, so OpenAI’s claim that Terra completes tasks more than three times faster than Astra and Sol is not independently corroborated by that benchmark.
OpenAI said 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, not an independent security finding. The available information does not detail the scope of that assessment or show how Terra performs in every cybersecurity scenario. Nor do the benchmark results establish whether Terra will be cheaper for a particular deployment after integration, review and error-handling costs.
Choosing a Setting for Each Workload
Users can test Terra in ChatGPT Work or Codex if they have an eligible Plus, Pro, Enterprise or Edu account, or access it through the API. The benchmark results point to low or medium effort as starting points for routine, high-volume tasks, followed by evaluation against real work. Organizations can compare cost per completed task, error rates and the amount of human review required before routing more work to the model.
For difficult coding or other demanding tasks, OpenAI’s own guidance is to consider Sol or Astra instead. Further comparisons will depend on additional independent speed measurements and results from real-world deployments. Until those are available, Terra’s release establishes a lower-priced option with promising cost-efficiency at selected settings, not proof that paying for a larger model is unnecessary.
Key Questions
Is GPT-6.1 Terra OpenAI’s most capable model?
No. OpenAI positioned Terra between Luna and its larger Sol and Astra models. Artificial Analysis scored Terra at 48.1 at maximum effort, below Sol at 51.8 and several other tested models.
How much does GPT-6.1 Terra cost?
OpenAI lists API pricing 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 evaluation set; that is a benchmark average, not a fixed charge for every task.
Which effort setting offers the best value?
Artificial Analysis identified low and medium effort as sitting on its cost-efficiency frontier. Low effort scored 36.2 at $0.055 per task in its evaluation, while maximum effort scored 48.1 at $0.40. The best setting for a given user will depend on the task and the quality required.
Does Terra make larger models unnecessary?
No. Terra costs less in the cited comparisons, but it also scored below several larger models. OpenAI says Sol and Astra remain better choices for its most complex coding tasks, so the results support choosing models by workload rather than replacing every larger model with Terra.
Source: ThorstenMeyerAI.com
