GPT-6.1 Terra: near-flagship scores for high-volume work
New models on Complete
GPT-6.1 Terra is out, and it's available for enterprise teams on Complete. It's OpenAI's new mid-tier model, and it scores within four points of GPT-6.1 Sol on the Artificial Analysis Intelligence Index at a lower cost to run.
OpenAI built Terra for visual reasoning, computer use and clear writing. For a business buyer, that means document-heavy work you run thousands of times a week: reviewing contracts, drafting reports, reading screenshots and forms, and working through long case files.
The numbers
Artificial Analysis runs every model through the same independent set of tests and publishes two figures that matter to a buyer: how well the model scores, and what it costs to run the full set.
Model | Intelligence Index (v4.3.2) | Cost to run the full index |
|---|---|---|
Claude Opus 5.5 | 57.6 | $8,708 |
Gemini 4 Argon | 52.6 | $2,101 |
GPT-6.1 Sol | 51.8 | $1,082 |
GPT-6.1 Terra | 48.1 | $909 |
DeepSeek V4.1 Flash 1017 | 45.1 | $445 |
Source: Artificial Analysis, October 2026.
Terra costs about a tenth of Claude Opus 5.5 and less than half of Gemini 4 Argon to complete the same work. It sits on the best-value line of the chart above, which means no model in this comparison scores higher for less money.
Artificial Analysis also reports that Terra costs less than a third of its predecessor, GPT-5.6 Terra, while scoring higher.
Where it performs
Coding tasks in a terminal: 55.6% on Terminal-Bench 4.0, according to Artificial Analysis.
Visual reasoning over charts, diagrams and documents: 84.7% on MMMU-Pro, according to Artificial Analysis.
Long inputs: a 1.05 million token context window, with up to 922,000 tokens of input. That covers a full contract set, a quarter's worth of customer tickets or a large codebase in a single request.
Adjustable effort: OpenAI lets you set reasoning effort from none up to max, so simple tasks run fast and hard ones get more thinking time.
What it costs
OpenAI lists Terra at $1 per million input tokens and $5 per million output tokens. Cached input drops to $0.05 per million, so repeated prompts and shared documents cost far less on the second pass. Batch and Flex processing run 50% below standard pricing.
Requests over 272,000 input tokens are billed at 2x the input rate and 1.5x the output rate for the whole request. Very long documents still fit, at a higher price per request.
What to watch
Terra answers from its own knowledge with a 38% hallucination rate on the Artificial Analysis knowledge test, which counts how often a model guesses wrong instead of declining. For factual work, ground it in your own documents and data. Its knowledge cutoff is June 1, 2026.
Terra reads text and images and writes text. OpenAI lists no audio or video input. Independent speed figures from Artificial Analysis are not published yet.
Run it on your own workflows
GPT-6.1 Terra is available for enterprise teams on Complete. Bring a real workload, such as a contract review queue, a reporting process or a ticket backlog, and see the results on your own data.
