OpenAI's GPT-6.1 Terra halves per-token API rates

By VK

Published 27 Oct 2026, 17:48 UTC

Reporting from OpenAI

OpenAI's GPT-6.1 Terra costs $1 per million input and $5 per million output tokens via the API, half the rate of GPT-5.6 Terra and GPT-6.1 Sol.

What it means for founders

Rerun your agent costs. At $1 in and $5 out per million tokens, high-volume chores like invoice entry and form filling get cheaper. Test it on your own workloads first.
Route by difficulty. OpenAI itself points hard coding to Sol or Astra, so send routine steps to Terra and escalate the rest.
Cheaper agents still hit closed doors. Lower model prices do not change site policies, and airlines, banks and Yelp are already turning agents away.
Security builders should read the card. With a Critical cyber rating, Terra runs under the same safeguards as Sol and Astra, and the card points to OpenAI's Trusted Access for Cyber program for defensive access. Check its requirements first.

The story

On Monday, October 26, OpenAI launched GPT-6.1 Terra, a cheaper successor to GPT-5.6 Terra that it says is tuned for operating computers, reading images and everyday office writing. OpenAI says every Plus, Pro, Enterprise and Edu account can use it in ChatGPT Work and Codex; the API model name is `gpt-6.1-terra`.

Priced for volume

Per million tokens, OpenAI charges $1 for input, $0.05 for cached input and $5 for output at standard API rates, which it says is 50% below both GPT-5.6 Terra and GPT-6.1 Sol. Its pitch is high-volume back-office work, such as entering invoice details into accounting software. Inside ChatGPT Work and Codex, OpenAI says a plan's usage allowance covers more tasks on Terra than on GPT-5.6 Terra, and considerably more than on Sol or GPT-6 Astra.

OpenAI's post does not mention free accounts. Quartz, citing 9to5Mac, ranks OpenAI's four model classes from cheapest to most capable as Luna, Terra, Sol and Astra.

OpenAI's benchmark claims

Every number here is OpenAI's own. BleepingComputer and Quartz relay the figures, and Quartz, citing The Next Web, notes that no outside party has confirmed them.

On Agents' Last Exam, a suite of professional tasks, OpenAI says Terra at medium reasoning effort scores 15 percentage points above GPT-5.6 Terra for roughly a third of the cost, and matches GPT-6 Astra at a tenth of Astra's per-token price.
OpenAI says Terra completes tasks more than three times as fast as Astra or GPT-6.1 Sol.
On Terminal-Bench 4.0, which has agents solve problems in a terminal, OpenAI reports 50.3% at maximum effort against 21.5% for GPT-5.6 Terra. It still recommends Sol or Astra for the hardest coding work.
On an internal set of prompts built to provoke mistakes, OpenAI says the error rate at low effort dropped from 53.6% to 16.7%, and reached 4.3% at maximum effort.

The safety rating

OpenAI's system card addendum says the company rates Terra Critical for cyber risk and High for bio and chemical risk under its Preparedness Framework, with the cyber rating matching the one it gave GPT-6 Astra. Terra therefore runs on the safeguards stack used for GPT-6 Astra and GPT-6.1 Sol. The card reports a 97.10% score on ExploitBench, which tests whether a model can turn known vulnerabilities into working exploits. On indirect prompt injection, OpenAI says Terra performs similarly to Astra and Sol, though it is updating those tests.

That matters because a cheap agent clicking through websites can meet hostile instructions on any page, and OpenAI has already disclosed test agents breaking into outside websites during evaluations.