🔍 Read the full analysis: GPT-6.1 Terra: A Potentially Cheaper Fit For Everyday AI Work on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
OpenAI released GPT-6.1 Terra on Oct. 26, 2026, positioning it as a lower-cost model for routine workplace and computer-based tasks. Artificial Analysis found its low and medium effort settings on the cost-efficiency frontier in its tests, while higher scores from larger models and uncertainty about real-world performance remain relevant.
OpenAI released GPT-6.1 Terra on Oct. 26, 2026, introducing a mid-tier model for routine work at API prices of $1 per million input tokens and $5 per million output tokens. Artificial Analysis’s launch-day evaluation found Terra’s low and medium reasoning settings among the more cost-efficient options it tested, though the model scored below larger systems on its Intelligence Index.
Terra is available through the API under the name gpt-6.1-terra and to eligible Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, according to OpenAI. The company describes it as a fast, efficient model for communication and computer-based tasks, including high-volume work such as reading invoices and entering information into accounting software. OpenAI says the same ChatGPT allowance covers substantially more tasks with Terra than with Sol or Astra.
The listed API price is 50% lower than GPT-5.6 Terra’s: $1/$5 per million input/output tokens, compared with $2/$12. Cached input costs $0.05 per million tokens. Artificial Analysis reported a score of 48.1 on its Intelligence Index for Terra at maximum effort, with an average cost of $0.40 per evaluation task. Its comparison lists GPT-6.1 Sol at 51.8 and $0.72 per task, and Claude Opus 5.5 at 57.6 and $5.98 per task.
OpenAI also reported its own benchmark results. On Agents’ Last Exam, it said Terra at medium effort scored 57.6% for $0.29 per task, compared with 42.6% and $0.84 for GPT-5.6 Terra. On Terminal-Bench 4.0, Terra at maximum effort solved 50.3% of tasks, while the published GPT-5.6 Terra result was 21.5%. These are company-reported measurements; they should not be treated as directly interchangeable with Artificial Analysis’s tests.
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 Cost Changes the Comparison
Terra’s launch highlights a distinction that matters for businesses and people paying for AI through usage limits: token price is not the same as cost per completed task. A cheaper model can use more tokens or require additional attempts, while a more expensive model may finish a task with fewer outputs. Artificial Analysis estimated Terra at low effort at $0.055 per task, below its $0.07 estimate for GPT-6 Luna at maximum effort, despite Luna’s lower listed token price. That result reflects the firm’s tested tasks and average costs, not a guarantee for every workload.
The six effort settings also make the product a question of matching capability to work. Artificial Analysis measured Terra from 33.5 index points at non-reasoning to 48.1 at maximum effort; maximum cost $0.40 per task, versus $0.055 at low effort. The added cost bought a higher score in that evaluation, but did not make Terra the top-scoring model. For organizations processing many routine requests, starting with low or medium effort and escalating harder cases could limit spending, though the right routing policy depends on the work and its error costs.
Terra is not the only lower-cost option in the comparison. Artificial Analysis listed open-weight MiMo-V2.6-Pro at 46.3 points and $0.13 per task, while DeepSeek V4.1 Flash scored 45.1 at $0.26. Those figures give buyers alternatives to examine, but benchmarks alone do not establish suitability, privacy terms, availability or total deployment cost.
Products Worth Considering
Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite
Apple iPad Air 13-inch (M4), 128GB, Wi-Fi 7, Space Gray
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
As an affiliate, we earn on qualifying purchases.
OpenAI’s Mid-Tier Model Lineup
OpenAI places Terra between its smaller GPT-6 Luna and larger GPT-6 Sol and Astra models. The company’s launch framing emphasizes everyday, high-volume computer work rather than the most complex tasks. OpenAI said Sol and Astra remain better choices for its hardest coding tasks, a qualification that limits any interpretation of Terra as a replacement across the model lineup.
Artificial Analysis published its evaluation on launch day. Its Intelligence Index combines ten evaluations spanning areas such as agentic office work, terminal tasks, science and hallucination, while its task-cost figures account for tokens used in those evaluations. The results are a snapshot: the source says model-page figures were retrieved on October 26, 2026, and benchmark scores depend on the tasks, settings and measurement methods used.
OpenAI further said Terra completes tasks more than three times faster than Astra and Sol, and that it meets the Critical threshold in cybersecurity under the company’s Preparedness Framework. Those statements are attributed to OpenAI; Artificial Analysis had not published Terra speed measurements at the time of the source report. OpenAI also reported lower error rates on its internal error-inducing prompts, but that internal test is not an independent measure of general factual accuracy.
Independent Results Have Limits
The launch-day figures do not show how Terra will perform across every company’s real workflows. Artificial Analysis’s task costs are averages across its evaluation set, and the supplied material does not provide the complete results for all six effort settings. It reports that low and medium sit on the cost-efficiency frontier, but the precise scores and costs for the remaining settings beyond the supplied entries are not fully available here.
Several important comparisons also rely on different sources. OpenAI’s speed, cybersecurity and internal error-rate statements are company claims, while the Intelligence Index and its average task costs come from Artificial Analysis. The source report says independent speed measurements were not yet available. Real-world latency, accuracy, data handling terms, regional availability and how the model behaves on a particular company’s tasks remain unclear.
Terra’s reported reduction in guessing on questions it could not answer also needs careful framing: Artificial Analysis found a 38% guess rate in its hallucination test, compared with 88% for GPT-5.6 Terra, according to the source. That is a result on a specific test, not evidence that Terra will avoid errors at the same rate in ordinary use.
Watch Real-World Cost Tests
The next useful evidence will come from independent testing across Terra’s supported effort settings, including comparable measures of accuracy, speed and cost per completed task. Companies evaluating the model can test it on representative workloads, starting with low or medium effort and checking whether the lower cost changes error rates or the need for human review.
OpenAI’s launch makes Terra available through its API and named paid ChatGPT plans, but the source material does not give a rollout schedule beyond availability on launch day. Buyers and users will need to compare Terra with larger OpenAI models and alternatives on their own tasks rather than infer suitability from token prices or a single benchmark score.
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. Cached input is listed at $0.05 per million tokens. The supplied source says those API rates are 50% lower than GPT-5.6 Terra’s.
Where can people use GPT-6.1 Terra?
According to OpenAI, Terra is available through the API as gpt-6.1-terra and in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users.
Does Terra outperform OpenAI’s larger models?
Not on the cited Artificial Analysis Intelligence Index comparison. Terra scored 48.1 at maximum effort, below Sol at 51.8 and Astra at 52.7 in the supplied figures. OpenAI says Sol and Astra remain better choices for its most complex coding tasks.
Which Terra setting is the best value?
Artificial Analysis said Terra’s low and medium effort settings were on the cost-efficiency frontier in its evaluation. The best setting for a specific use depends on task difficulty, acceptable error rates and the cost of human review.
Are the reported speed and safety results independently verified?
The source attributes Terra’s speed and cybersecurity classification to OpenAI. It says Artificial Analysis had not published speed measurements when the report was prepared. The provided material does not establish independent verification of those claims.
Source: ThorstenMeyerAI.com
