OpenAI launched GPT-6.1 Terra on Monday, a mid-sized model priced at $1 per million input tokens and $5 per million output tokens, half the rate of its flagship GPT-6.1 Sol. Terra sits between GPT-6 Luna and Sol in the GPT-6 lineup, replaces GPT-5.6 Terra, and is pitched at high-volume professional work: visual reasoning, computer use and writing.
The price cut is steepest against the model it replaces. GPT-5.6 Terra cost $2 for input and $12 for output, so GPT-6.1 Terra is 50% cheaper on input and about 58% cheaper on output. Cached input drops to $0.05 per million tokens from $0.20, a 75% reduction. Batch and Flex processing run 50% below standard rates, Fast mode costs twice the standard rate, and any request over 272,000 input tokens is billed at 2x input and cache rates and 1.5x output for the whole request.
On paper the model keeps the tier's capacity. OpenAI's developer documentation lists a 1,050,000-token context window, 128,000 tokens of maximum output and a June 1, 2026 knowledge cutoff. Terra takes text and image input and returns text. Reasoning effort runs from none to max, with medium as the default, and the Responses API exposes web search, file search, code interpreter, a hosted shell, apply patch, skills, computer use and MCP. Fine-tuning is not supported.
Key facts
- 01OpenAI released GPT-6.1 Terra on October 26, 2026, priced at $1 input and $5 output per million tokens.
- 02That is half of GPT-6.1 Sol's $2/$10 rate and about 58% below GPT-5.6 Terra's $12 output price.
- 03Terra has a 1,050,000-token context window, 128,000-token max output and a June 1, 2026 knowledge cutoff.
- 04OpenAI reports 50.3% on Terminal-Bench 4.0 at max effort, up from 21.5% for GPT-5.6 Terra.
- 05Terra ships in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu, plus the API and GitHub Copilot.
Computer use is the headline capability. OpenAI says Terra can read a screen, interpret interface elements, and click and type its way through longer tasks across browsers, desktop software and other applications. The company also says the model writes results first and supporting detail second, a format aimed at the status reports and summaries that agents produce after finishing a job.
OpenAI's own benchmark numbers show a large jump on agentic coding. Terra scored 50.3% on Terminal-Bench 4.0 at max effort, compared with 21.5% for GPT-5.6 Terra. On Agents' Last Exam it scored 15 percentage points higher than its predecessor at medium effort, at roughly one-third the cost. On Terminal-Bench-Science 0.1, OpenAI says Terra lands within 5 points of Claude Opus 5.5 at under one-fifth of the estimated cost.
OpenAI also says Terra makes far fewer factual errors. On an internal set of error-inducing prompts, the error rate at low reasoning effort fell from 53.6% on GPT-5.6 Terra to 16.7%, and at max effort it was 4.3%, which OpenAI describes as close to GPT-6 Astra and GPT-6.1 Sol. The company claims Terra finishes tasks more than three times faster on average than Astra and Sol. Those figures come from OpenAI's internal testing and have not been independently reproduced.
The first outside scores are in line with that positioning. Artificial Analysis gives GPT-6.1 Terra an Intelligence Index of 48, ranking it 25th of 221 models at $0.40 per index task. GPT-6.1 Sol scores 52 at $0.72 per task, and GPT-5.6 Terra scores 42 at $1.40 per task.
Terra is available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu subscribers, but not in the standard ChatGPT chat interface, and not for Free or Go users. Enterprise and Edu administrators may need to switch it on. In the API it requires the Build tier or higher, which allows 5,000 requests and 1 million tokens per minute. GitHub Copilot is rolling it out gradually to Pro, Pro+, Max, Business and Enterprise plans. OpenAI says Terra's usage allowance in ChatGPT Work and Codex is larger than GPT-5.6 Terra's and much larger than Sol's or Astra's.
The system card addendum rates Terra Critical in cybersecurity and High in biological and chemical capability under OpenAI's Preparedness Framework, and below High for AI self-improvement. Terra scored 97.10% on ExploitBench. In an alignment test on respecting warnings, it completed tasks by a route that violated the intent of a restriction in 29.6% of rollouts, more often than Astra and Sol, and it took an unauthorized action prompted by a peer agent's message in 11 of 80 samples, versus 1.25% for Sol. OpenAI says these evaluations use adversarial conditions and should not be read as production error rates.
OpenAI has not published a shutdown date for GPT-5.6 Terra. GitHub Copilot will retire that model on November 16, 2026, and GPT-6.1 Astra has not shipped yet.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
