🔍 Read the full analysis: Should You Choose GPT-6.1 Terra For Your Next AI Task? 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 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; low and medium settings offered stronger cost efficiency, while harder tasks may call for larger models.
OpenAI released GPT-6.1 Terra on October 26 as a lower-cost model for workplace and computer-based tasks, placing it between its smaller Luna model and larger Sol and Astra models. Independent evaluator Artificial Analysis measured Terra at $0.40 per task and a 48.1 Intelligence Index score at maximum effort, suggesting its value depends on the task and reasoning setting rather than its token price alone.
OpenAI lists API pricing at $1 per million input tokens and $5 per million output tokens, compared with $2 and $12 for GPT-5.6 Terra. Cached input costs $0.05 per million tokens. The model is available through ChatGPT Work, Codex and the API; OpenAI says paid ChatGPT plans can complete substantially more tasks with the same allowance than when using Sol or Astra.
Artificial Analysis evaluates models across ten tasks and includes the cost of tokens used in completing them. At maximum effort, its figures put Terra at 48.1 points and $0.40 per task. GPT-6.1 Sol scored 51.8 at $0.72, while Claude Opus 5.5 scored 57.6 at $5.98. These are averages across the evaluator’s tasks, not a guaranteed price or result for any particular user’s work.
Terra offers six effort settings. Artificial Analysis reported scores ranging from 33.5 to 48.1 and per-task costs from $0.055 at low effort to $0.40 at maximum. Its measurements place low and medium effort on the cost-efficiency frontier: in that comparison, no tested model offered both a lower cost and a higher score. Maximum effort added 12 points over low, but cost about 7.4 times as much. The non-reasoning setting cost 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
Task Costs Change the Model Choice
Terra’s pricing matters most for organizations that run repetitive, high-volume workflows, where a small difference per completed task can affect total operating costs. OpenAI gives reading invoices and entering them into accounting software as an example. A lower token rate may also help users stretch a fixed allowance, but the useful comparison is the cost and quality of finishing the work, not the input price in isolation.
That distinction explains why a model with a higher price per token can sometimes cost less for a task: it may use fewer tokens to produce an answer. Artificial Analysis found Terra at low effort cost less per task than GPT-6 Luna despite Luna’s lower per-token price. For buyers, the practical question is whether Terra completes their specific workflow accurately and with acceptable review needs. The benchmarks offer a starting point, not a substitute for testing on real workloads.
Terra is not the top-scoring choice in the cited comparison. Claude Opus 5.5, Kimi K3.1 Pro and Gemini 4 Argon all scored higher on Artificial Analysis’s index. Xiaomi’s open-weight MiMo-V2.6-Pro scored 46.3 for $0.13 per task, less than Terra’s maximum-effort cost. Those results show that Terra competes on a combination of performance and cost; they do not establish that it is the cheapest or best option overall.
Products Worth Considering
Dell UltraSharp 32 U3226QER 4K Thunderbolt Hub Monitor
Dell UltraSharp 27 Thunderbolt Hub Monitor (U2726DER)
Dell XPS 13 9350 AI Professional Laptop 13.4 120Hz FHD+ Display
As an affiliate, we earn on qualifying purchases.
OpenAI’s Mid-Tier Model Position
OpenAI describes Terra as a “fast, efficient model” for communication and computer-based tasks. Its release places Terra between Luna and the more capable Sol and Astra models, with the company pitching it for routine work rather than the hardest problems. OpenAI says Terra completes tasks more than three times faster than Sol and Astra, but Artificial Analysis had not published speed measurements for Terra at the time of the source report. That speed comparison remains the company’s claim.
OpenAI also reported improvements on several of its own tests. On Agents’ Last Exam, it said Terra at medium effort scored 57.6%, compared with 42.6% for GPT-5.6 Terra, at a reported cost of $0.29 rather than $0.84 per task. On Terminal-Bench 4.0, OpenAI reported a 50.3% task success rate at maximum effort, compared with 21.5% for GPT-5.6 Terra. The company cautions that Sol and Astra remain better choices for its most complex coding tasks.
OpenAI’s internal error-prompt test also showed lower reported error rates: 16.7% at low effort and 4.3% at maximum, down from 53.6% and 22.3% respectively for GPT-5.6 Terra. Separately, Artificial Analysis’s hallucination test found Terra guessed on 38% of questions it could not answer, compared with 88% for its predecessor. These results come from different tests and should not be treated as a single, universal measure of factual reliability.
How Terra Performs in Real Work
The available results do not establish how Terra will perform on every customer’s documents, software or coding tasks. Benchmark scores depend on the tests, settings and token use, and the task-cost figures are averages rather than fixed charges. OpenAI’s internal measurements and Artificial Analysis’s evaluation also use separate methodologies, so their percentages and scores cannot be directly merged.
Speed is another open point: OpenAI claims Terra is more than three times faster than Sol and Astra, but the independent evaluator had not published speed results when the report was written. The source material also does not provide enough detail to verify how results vary across all six effort levels beyond the reported range and selected settings. Actual performance, latency and cost may differ by workload.
It is also unclear how often users will need human review, or whether Terra’s lower reported error and hallucination rates will hold in production. OpenAI says Terra meets the Critical threshold in its cybersecurity Preparedness Framework and retains safeguards developed for Astra; that classification is the company’s own assessment, not an independent safety finding in the cited material.
Test Workloads Before Routing
For teams considering Terra, the next practical step is to run representative tasks at low and medium effort, then compare completion quality, review time and total cost against current models. Work that needs stronger reasoning or handles more complex coding can be routed to Sol or Astra, consistent with OpenAI’s guidance. The source material does not identify a later benchmark release or a new Terra feature date.
Individual users can try Terra through the listed ChatGPT Work and Codex access or the API, subject to their plan and configuration. They should judge results on their own tasks rather than assume the benchmark average predicts every outcome. Further independent speed measurements and broader real-world results would help clarify whether Terra’s cost advantage holds across different workloads.
Key Questions
Is GPT-6.1 Terra OpenAI’s most capable model?
No. OpenAI positions Terra between Luna and its larger Sol and Astra models. Artificial Analysis also ranked several other tested models above Terra by Intelligence Index score.
How much does GPT-6.1 Terra cost?
OpenAI lists API rates of $1 per million input tokens and $5 per million output tokens, with cached input at $0.05 per million. Artificial Analysis measured an average task cost of $0.40 at maximum effort, but actual task costs vary.
Which effort setting is the best value?
In Artificial Analysis’s comparison, low and medium effort were on the cost-efficiency frontier. Low effort cost $0.055 per task in its measurements; the right setting depends on the task’s quality requirements.
Should I use Terra for complex coding?
OpenAI says Sol and Astra remain better choices for its most complex coding tasks. Terra posted improved results over its predecessor on Terminal-Bench 4.0, but teams should test their own coding workloads before choosing a model.
Are Terra’s speed claims independently verified?
Not in the cited evaluation at launch. OpenAI said Terra is more than three times faster than Sol and Astra, while Artificial Analysis had not yet published speed measurements for Terra.
Source: ThorstenMeyerAI.com
