GPT-6.1 Terra costs half as much as Sol and carries Astra's Critical cyber rating

OpenAI has released GPT-6.1 Terra, the model that sits between Luna and Sol in its current lineup, pitched at screen reading, computer use and everyday writing. It is available in the API as gpt-6.1-terra and in ChatGPT Work and Codex.

The price is the easiest part to check. On the pricing page, standard short-context rates are $1 per million input tokens, $0.05 for cached input and $5 for output. That is exactly half of GPT-6.1 Sol ($2 / $0.10 / $10) and a tenth of GPT-6 Astra ($10 / $50). Against GPT-5.6 Terra ($2 / $12), input is half price and output drops by more than half. Prompts over 272K tokens are billed at 2x the input and cache rates and 1.5x output for the whole request. Batch and Flex cost half the standard rate, and Fast mode costs double.

The model page lists a 1,050,000-token context window, with input capped at 922,000 tokens and output at 128,000. The knowledge cutoff is June 1, 2026. Reasoning effort runs from none to max. Chat Completions is supported, but tool calling requires the Responses API, so the endpoint choice matters for agent builders.

What OpenAI claims

Every performance figure in the launch post comes from OpenAI. On Agents' Last Exam, a professional-workflow benchmark built by Berkeley RDI but run here by OpenAI, the company says Terra at medium effort scores 15 points above GPT-5.6 Terra at about a third of the cost and performs on par with Astra. On Terminal-Bench 4.0 at max effort, it reports 50.3% against GPT-5.6 Terra's 21.5%. OpenAI also says Sol and Astra remain the better choice for the hardest coding work. The hallucination figures use an internal set of error-inducing prompts. At low effort, OpenAI says the error rate fell from 53.6% with GPT-5.6 Terra to 16.7%.

OpenAI's Thibault Sottiaux puts the subscription side as "2X more usage than 6.1 Sol across API and subscriptions". The API half of that matches the price sheet.

The cyber rating came along too

The more interesting document is the system card addendum. OpenAI treats Terra as Critical in cybersecurity under its Preparedness Framework, the same top tier it assigned to Astra, and as High in biology and chemistry. Terra gets the same safeguards stack as Sol and Astra.

The numbers behind the cyber rating aren't simply a smaller Astra. On an internal set of exploits for vulnerabilities disclosed between June and August 2026, Terra reached arbitrary code execution in 24.5% of attempts. That beats GPT-6.1 Sol's 21.5% and trails Astra's 31.5%. It is also a jump from the 5.5% OpenAI reports for GPT-6 Sol. On Sandbox Bench, which targets runtimes, parsers and egress proxies, Terra broke 5 of 22 targets, against 9 for Sol and 10 for Astra. The card labels Sol's result 41.9%, though 9 of 22 works out to 40.9%.

In practice, the capability tier OpenAI gave Astra in September is now on sale at a tenth of Astra's per-token price. Whether the safeguards hold up is now a bigger question, because many more requests will run through them.

The alignment section is mixed. In OpenAI's simulation of internal Codex traffic, Terra drew 22 severity-3-or-higher flags across 48,690 tasks, slightly fewer than GPT-6.1 Sol (24) and Astra (26). In other tests it does worse than Sol. Terra completed a blocked task by a route that "violated the spirit of the restriction" in 29.6% of max-effort rollouts, more often than Astra or GPT-6.1 Sol, though less often than GPT-6 Sol. In a scenario with messages from other agents, it performed the specified unauthorized action in 11 of 80 samples, against 1 for GPT-6.1 Sol. It also showed signs of recognizing it was being evaluated in 9.88% of simulated Codex trajectories, the highest of the five models compared. OpenAI notes that models increasingly recognize evaluations, which makes some of these tests harder to read.

None of this has been checked independently yet. For now, the safest reading is that the prices are confirmed, the benchmark scores are OpenAI's claims, and the system card shows a model that is cheaper than Sol without being uniformly safer.