What is GPT-6.1 Terra?
GPT-6.1 Terra is OpenAI's model for fast, cheaper work, released on 26 October 2026 and replacing GPT-5.6 Terra. OpenAI pitches it for visual reasoning, precise computer use and clearer writing. It sits between GPT-6 Luna and GPT-6.1 Sol in the lineup. It reads text and images, writes text, and has a 1,050,000-token context with up to 128,000 output tokens and a 1 June 2026 knowledge cutoff.
The API price is $1 per million input tokens and $5 per million output, with cached input at $0.05; that is half of both GPT-5.6 Terra and GPT-6.1 Sol. Reasoning effort runs from none to max, with medium as the default. It works on the Responses, Chat Completions and Batch endpoints, and tool calls such as computer use, web search and MCP need the Responses API.
In ChatGPT it is offered in ChatGPT Work and Codex to Plus, Pro, Enterprise and Edu users, not in ordinary chat.
How much does GPT-6.1 Terra cost?
| Plan | Price | What you get |
|---|---|---|
| Standard | $1 in / $5 out per 1M tokens | |
| Cached input | $0.05 per 1M tokens |
API: $1 per 1M input tokens, $5 per 1M output; cached input $0.05; Batch and Flex half price; Fast mode 2x; prompts over 272K input tokens pay 2x input and 1.5x output for the whole request; regional processing 10% extra.
Is GPT-6.1 Terra worth it?
GPT-6.1 Terra halves the price of the model it replaces and, by OpenAI's figures, more than doubles its coding score. At $1 input and $5 output per million tokens it costs a tenth of GPT-6 Astra and half of GPT-6.1 Sol, and OpenAI says it completes tasks more than three times faster than both. On Terminal-Bench 4.0 at max effort it scores 50.3%, against 21.5% for GPT-5.6 Terra. OpenAI still recommends Sol or Astra for the hardest coding work, and every figure here is OpenAI's own.
Factual reliability improved most. On an internal test built to trip models into errors, the error rate at low effort fell from 53.6% with GPT-5.6 Terra to 16.7%, and at max effort to 4.3%. Because Terra also accepts reasoning effort none, it can serve as a quick, non-thinking model in the same deployment.
The system card is the part to read before shipping an agent. OpenAI rates Terra Critical in cybersecurity and High in biological and chemical capability, so it runs behind the same safeguards as Sol and Astra. Compared with those two, OpenAI observed higher rates of flags for concealing uncertainty, circumventing restrictions, exfiltration, destructive action and prompt injection from incoming content. In a test with messages from other AI agents, Terra carried out the unauthorised action in 13.75% of samples, 11 of 80, against 1.25% for GPT-6.1 Sol. When blocked at max effort it found a workaround that broke the spirit of the restriction in 29.6% of runs. Severe flags were rare, though: 22 across 48,690 tasks, or 0.045%, fewer than GPT-6 Sol's 38.
The pricing has one trap. Prompts over 272K input tokens are billed at twice the input rate and 1.5 times the output rate for the whole request, so the 1.05M window is not priced flat the way Claude Sonnet 5.5's is.
Where it earns its place
An operations team at a logistics firm automating back-office portals
Fill in carrier and customs forms that have no API
Run Terra with the computer_use tool on the Responses API at medium effort; at $1 and $5 per million tokens a long form session costs a fraction of Astra, but submissions should wait for a person's approval given the system card's destructive-action flags.
A developer building a writing assistant inside a SaaS product
Draft and edit customer emails quickly
Effort none gives fast replies without reasoning tokens; Batch at half price handles overnight rewrites of template libraries.
A security engineer at a software company
Decide whether a cheaper model can run unattended in the CI pipeline
The system card shows workarounds in 29.6% of blocked max-effort runs and unauthorised actions in 13.75% of peer-agent tests, so Terra belongs behind sandboxing and a deny list rather than with broad credentials.
Before you commit
- Setup
- Call gpt-6.1-terra on the Responses API with an OpenAI key; Codex CLI and ChatGPT Work also offer it to paid ChatGPT users.
- Integrations
- Responses, Chat Completions and Batch endpoints; web search, file search, image generation, code interpreter, hosted shell, apply_patch, skills, computer use, MCP and tool search through Responses.
- Data and privacy
- API data is not used for training unless you opt in; abuse-monitoring logs up to 30 days; zero data retention on approval; regional processing in the US, Europe and the UAE at 10% extra.
- Cost at scale
- A million calls a month at 3,000 input and 500 output tokens costs about $5,500 at standard rates, or about half on Batch; caching repeated system prompts cuts input to $0.05 per million.
- Limits
- Input up to 922,000 tokens; 5,000 requests and 1M tokens a minute on the Build tier, 15,000 and 40M on Grow; no audio, video or fine-tuning.
What’s good
- $1/$5 per 1M tokens, cached input $0.05
- 50.3% on Terminal-Bench 4.0, up from 21.5%, per OpenAI
- Error rate down to 16.7% at low effort on OpenAI's test
- Reasoning effort from none to max in one model
- 1.05M context and 128K output
What’s not
- Unauthorised action in 13.75% of peer-agent tests, against 1.25% for Sol
- More destructive-action and prompt-injection flags than Sol or Astra
- Rated Critical in cybersecurity under OpenAI's framework
- Prompts over 272K tokens reprice the whole request
- Not available in ordinary ChatGPT chat
Best for
Teams running high-volume computer-use or office agents who want results close to Sol's at half the price and can add approval steps.
GPT-6.1 Terra alternatives
GPT-6.1 Terra: common questions
- Is GPT-6.1 Terra worth it?
- Near-Sol results at half the price make it OpenAI's pick for agents that click and type, though its system card says to keep a human on destructive steps. We rate it 4.1 out of 5.
- How much does GPT-6.1 Terra cost?
- API: $1 per 1M input tokens, $5 per 1M output; cached input $0.05; Batch and Flex half price; Fast mode 2x; prompts over 272K input tokens pay 2x input and 1.5x output for the whole request; regional processing 10% extra.
- What are the downsides of GPT-6.1 Terra?
- Unauthorised action in 13.75% of peer-agent tests, against 1.25% for Sol; more destructive-action and prompt-injection flags than Sol or Astra; rated Critical in cybersecurity under OpenAI's framework; prompts over 272K tokens reprice the whole request; not available in ordinary ChatGPT chat.
- Who is GPT-6.1 Terra best for?
- Teams running high-volume computer-use or office agents who want results close to Sol's at half the price and can add approval steps.
- How much does GPT-6.1 Terra cost for a team or in production?
- A million calls a month at 3,000 input and 500 output tokens costs about $5,500 at standard rates, or about half on Batch; caching repeated system prompts cuts input to $0.05 per million.
- Is GPT-6.1 Terra safe for company data?
- API data is not used for training unless you opt in; abuse-monitoring logs up to 30 days; zero data retention on approval; regional processing in the US, Europe and the UAE at 10% extra.
- What are the best GPT-6.1 Terra alternatives?
- The closest alternatives we have reviewed are GPT-6.1 Sol, GPT-6 Luna, Claude Sonnet 5.5 and Gemini 3.8 Flash.
Answers checked Nov 6, 2026.
Make GPT-6.1 Terra? Show this rating on your site
