OpenAI’s GPT-6.1 Terra: $1/$5 per million tokens, and benchmarks from OpenAI’s own tests

OpenAI released GPT-6.1 Terra on 26 October, and its pitch to developers is price. The model documentation lists it at $1 per million input tokens and $5 per million output tokens, and OpenAI says those standard API prices are 50% lower than those of GPT-5.6 Terra and GPT-6.1 Sol. OpenAI describes Terra as “a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing”, aimed at work like filling out forms and interpreting technical drawings.

In ChatGPT, the announcement lists access for Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. OpenAI’s release notes also list Business plans, which the announcement doesn’t, and state that “Terra is available in Work and Codex, not Chat.” Neither mentions the Free tier. Developers can call the model through the API as gpt-6.1-terra.

The headline rate comes with conditions. Once a prompt goes past 272K input tokens, the docs say input and cache rates double and output costs 1.5x, and that applies to the full request, not just the tokens above the threshold. That matters because the context window is large: 1,050,000 tokens, with up to 922,000 tokens of input and 128,000 of output. Cached input costs $0.05 per million tokens and cache writes $1.25 per million with 30-minute retention. Batch and Flex prices are 50% lower than Standard, Fast mode is 2x Standard, and regional processing adds 10% where it’s available.

The model takes text and images and returns text. It doesn’t handle audio or video, and its knowledge cutoff is 1 June 2026. The reasoning.effort setting runs from none through low, medium (the default), high and xhigh to max. Fine-tuning isn’t supported.

Integrators should note one constraint. Chat Completions is a supported endpoint, but the docs state that “tool calling requires the Responses API”. That is also where the hosted tools live, including computer use, web search, file search, code interpreter and MCP. If you’re wiring Terra into an agent through MCP, our explainer on what MCP standardises and what the host must decide covers the parts the protocol leaves to you.

The case that Terra is worth switching to rests on OpenAI’s own tests, and we haven’t seen independent evaluations. On Agents’ Last Exam, a benchmark of professional computer-use tasks, OpenAI says Terra at medium reasoning effort scores 15 percentage points higher than GPT-5.6 Terra at about one-third of the cost, and performs on par with GPT-6 Astra at one-tenth of its price per token. OpenAI also says Terra completes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol.

On Terminal-Bench 4.0, OpenAI reports that Terra at maximum effort solves 50.3% of command-line coding tasks, against GPT-5.6 Terra’s published score of 21.5%. The company itself says GPT-6.1 Sol and GPT-6 Astra “remain the better choice” for the most complex coding work.

The largest claimed gain is the hardest to check. On OpenAI’s internal benchmark of error-inducing prompts, the company says Terra’s error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7%, and at maximum effort fell by 18 percentage points to 4.3%. Outsiders can’t inspect that test set.

Terra is also on Amazon Bedrock, where AWS announced general availability the same day. AWS says the model can be called through Bedrock APIs, and that ChatGPT Work and Codex can be configured to use it there. AWS’s own price claim is that input and output token prices are “at least 50% lower” than those of GPT-6.1 Sol and GPT-5.6 Terra, but the post doesn’t say whether it means OpenAI’s prices or Bedrock’s. It also describes “a context window of up to 1 million input tokens”, which doesn’t match the 922,000-token input limit in OpenAI’s docs, so check the limits on the platform you use. Earlier this month we covered a similar launch-day arrival on Bedrock for Claude structured outputs.

On safety, OpenAI’s system card addendum says the company is “treating GPT-6.1 Terra as Critical capability in Cybersecurity, High capability in the Biological and Chemical domain, and below the High threshold in AI Self-Improvement.” Under OpenAI’s Preparedness Framework, Critical cyber capability means a model that can “identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems without human intervention” or “devise and execute end-to-end novel strategies for cyberattacks against hardened targets given only a high level desired goal.” Terra uses the same safeguards stack as GPT-6.1 Sol and GPT-6 Astra, “with probes updated to be compatible with the model”, and the fuller description of those safeguards sits in the GPT-6 Astra system card.

The Bedrock model card does list prices. Global cross-Region inference costs $1 per million input tokens and $5 per million output, the same as OpenAI, while in-Region and geographic inference cost $1.10 and $5.50, which includes a 10% fee over OpenAI’s rates. In-Region inference is listed only in us-east-1 (N. Virginia) on the bedrock-mantle endpoint, and geographic inference only from six US and Canadian Regions. Global inference is listed from 31 Regions across North America, Europe, Asia-Pacific, Israel, Mexico, South Africa and Brazil.