OpenAI released GPT-6.1 Terra on October 26, a fast mid-tier model built for the computer-based work that fills a normal workday: reading documents, filling in forms, drafting proposals, interpreting technical drawings, matching product photos to catalogue entries and editing videos. The model is available now to ChatGPT Plus, Pro, Enterprise and Edu users inside ChatGPT Work and Codex, and to developers through the API under the name gpt-6.1-terra. The pitch is not raw power but unit economics — OpenAI prices the standard API at 1 dollar per million input tokens, 0.05 dollars per million cached input tokens and 5 dollars per million output tokens, exactly 50 percent below both GPT-5.6 Terra and the more powerful GPT-6.1 Sol.
Performance doubles on coding while staying deliberately mid-tier
On OpenAI's own benchmarks, Terra solves 50.3 percent of tasks on Terminal-Bench 4.0 at maximum reasoning effort, more than double the 21.5 percent published for GPT-5.6 Terra. On Agents' Last Exam, a benchmark of professional computer-use tasks, Terra at medium effort scores 15 points higher than GPT-5.6 Terra at about one-third the cost, and performs on par with GPT-6 Astra at one-tenth the price per token. The company says tasks finish more than three times faster than on Astra or Sol, and that at maximum effort Terra reaches roughly nine times lower cost per task than Claude Opus 5.5 while approaching its score. Terra handles just over one million tokens of context and nearly matches the flagship models on the MMMU-Pro visual reasoning test at much lower cost.
OpenAI has also pushed on factuality: on its internal set of prompts designed to induce errors, Terra's error rate at low reasoning effort fell from 53.6 percent with GPT-5.6 Terra to 16.7 percent, and at maximum effort it dropped 18 points to 4.3 percent, nearly matching Astra and Sol. The company remains explicit that for the hardest coding work, GPT-6.1 Sol and GPT-6 Astra are still the better choice — a boundary statement that positions Terra as the volume workhorse rather than the frontier flag-bearer.
A Critical safety rating shadows the price cut
Under OpenAI's Preparedness Framework, GPT-6.1 Terra meets the Critical threshold in cybersecurity capability — the same top risk tier — while keeping the protections built for GPT-6 Astra. The company reports stronger resistance to prompt injection, the hidden-instruction attack that tries to pull an agent away from its task, and ships a published system card with the release. The combination of cheaper access and high offensive-cyber capability is raising eyebrows: making a Critical-rated model significantly more affordable widens the surface for misuse even as it lowers the barrier for legitimate automation.
Independent early readings are mixed. Artificial Analysis puts Terra's intelligence index at 48 at maximum effort, up 6 points from the previous generation but still below GPT-6.1 Sol's 52 and Claude Opus 5.5's 58 — while its cost per task fell from 1.40 dollars to 0.40 dollars, a cut of more than 70 percent. Some users grumbled that the market wanted a model stronger than Opus 5.5, not merely a faster and cheaper one, according to TMTPost. For businesses running high-volume agents — invoice entry, batch form filling, screen-watching processes — the math changes overnight, and the center of gravity in model competition is visibly shifting from leaderboard scores to unit economics.
Sources: OpenAI, TMTPost, TechNewsReel, AI Parabellum, The New Stack, Aivy, threerouter
Top comments (1)
Hope it can beat Opus 5.5.